libxo - A Library for Generating Text, XML, JSON, and HTML Output
You want to prepare for the future, but you need to live in the present. You'd love a flying car, but need to get to work today. You want to support features like XML, JSON, and HTML rendering to allow integration with NETCONF, REST, and web browsers, but you need to make text output for command line users.
And you don't want multiple code paths that can't help but get out of sync:
/* None of this "if (xml) {... } else {...}" logic */
if (xml) {
/* some code to make xml*/
} else {
/* other code to make text */
/* oops forgot to add something on both clauses! */
}
/* And ifdefs are right out. */
#ifdef MAKE_XML
/* icky */
#else
/* pooh */
#endif
But you'd really, really like all the fancy features that modern encoding formats can provide. libxo can help.
The libxo library allows an application to generate text, XML, JSON, and HTML output using a common set of function calls. The application decides at run time which output style should be produced. The application calls a function "xo_emit" to product output that is described in a format string. A "field descriptor" tells libxo what the field is and what it means. Each field descriptor is placed in braces with a printf-like format string (Format Strings):
Each field can have a role, with the 'value' role being the default, and the role tells libxo how and when to render that field (see Field Roles for details). Modifiers change how the field is rendered in different output styles (see Field Modifiers for details. Output can then be generated in various style, using the "‑‑libxo" option:
We are following the branching scheme from http://nvie.com/posts/a-successful-git-branching-model/ which means we will do development under the "develop" branch, and release from the "master" branch. To clone a developer tree, run the following command:
You can retrieve the source for libxo in two ways:
A) Use a "distfile" for a specific release. We use github to maintain our releases. Visit github release page (https://github.com/Juniper/libxo/releases) to see the list of releases. To download the latest, look for the release with the green "Latest release" button and the green "libxo‑RELEASE.tar.gz" button under that section.
After downloading that release's distfile, untar it as follows:
tar -zxf libxo-RELEASE.tar.gz
cd libxo-RELEASE
[Note: for Solaris users, your "tar" command lacks the "‑z" flag, so you'll need to substitute "gzip -dc "file" | tar xf -" instead of "tar -zxf "file"".]
B) Use the current build from github. This gives you the most recent source code, which might be less stable than a specific release. To build libxo from the git repo:
git clone https://github.com/Juniper/libxo.git
cd libxo
_BE AWARE_: The github repository does _not_ contain the files generated by "autoreconf", with the notable exception of the "m4" directory. Since these files (depcomp, configure, missing, install-sh, etc) are generated files, we keep them out of the source code repository.
This means that if you download the a release distfile, these files will be ready and you'll just need to run "configure", but if you download the source code from svn, then you'll need to run "autoreconf" by hand. This step is done for you by the "setup.sh" script, described in the next section.
To build libxo, you'll need to set up the build, run the "configure" script, run the "make" command, and run the regression tests.
The following is a summary of the commands needed. These commands are explained in detail in the rest of this section.
sh bin/setup.sh
cd build
../configure
make
make test
sudo make install
The following sections will walk through each of these steps with additional details and options, but the above directions should be all that's needed.
Configure (and autoconf in general) provides a means of building software in diverse environments. Our configure script supports a set of options that can be used to adjust to your operating environment. Use "configure --help" to view these options.
We use the "build" directory to keep object files and generated files away from the source tree.
To run the configure script, change into the "build" directory, and run the "configure" script. Add any required options to the "../configure" command line.
cd build
../configure
Expect to see the "configure" script generate the following error:
/usr/bin/rm: cannot remove `libtoolT': No such file or directory
This error is harmless and can be safely ignored.
By default, libxo installs architecture-independent files, including extension library files, in the /usr/local directories. To specify an installation prefix other than /usr/local for all installation files, include the --prefix=prefix option and specify an alternate location. To install just the extension library files in a different, user-defined location, include the --with-extensions-dir=dir option and specify the location where the extension libraries will live.
libxo includes a set of regression tests that can be run to ensure the software is working properly. These test are optional, but will help determine if there are any issues running libxo on your machine. To run the regression tests:
Once the software is built, you'll need to install libxo using the "make install" command. If you are the root user, or the owner of the installation directory, simply issue the command:
make install
If you are not the "root" user and are using the "sudo" package, use:
sudo make install
Verify the installation by viewing the output of "xo --version":
% xo --version
libxo version 0.3.5-git-develop
xo version 0.3.5-git-develop
Most unix commands emit text output aimed at humans. It is designed to be parsed and understood by a user. Humans are gifted at extracting details and pattern matching in such output. Often programmers need to extract information from this human-oriented output. Programmers use tools like grep, awk, and regular expressions to ferret out the pieces of information they need. Such solutions are fragile and require maintenance when output contents change or evolve, along with testing and validation.
Modern tool developers favor encoding schemes like XML and JSON, which allow trivial parsing and extraction of data. Such formats are simple, well understood, hierarchical, easily parsed, and often integrate easier with common tools and environments. Changes to content can be done in ways that do not break existing users of the data, which can reduce maintenance costs and increase feature velocity.
In addition, modern reality means that more output ends up in web browsers than in terminals, making HTML output valuable.
libxo allows a single set of function calls in source code to generate traditional text output, as well as XML and JSON formatted data. HTML can also be generated; "<div>" elements surround the traditional text output, with attributes that detail how to render the data.
A single libxo function call in source code is all that's required:
There are four encoding styles supported by libxo:
TEXT output can be display on a terminal session, allowing compatibility with traditional command line usage.
XML output is suitable for tools like XPath and protocols like NETCONF.
JSON output can be used for RESTful APIs and integration with languages like Javascript and Python.
HTML can be matched with a small CSS file to permit rendering in any HTML5 browser.
In general, XML and JSON are suitable for encoding data, while TEXT is suited for terminal output and HTML is suited for display in a web browser (see xohtml).
Most traditional programs generate text output on standard output, with contents like:
36 ./src
40 ./bin
90 .
In this example (taken from du source code), the code to generate this data might look like:
printf("%d\t%s\n", num_blocks, path);
Simple, direct, obvious. But it's only making text output. Imagine using a single code path to make TEXT, XML, JSON or HTML, deciding at run time which to generate.
libxo expands on the idea of printf format strings to make a single format containing instructions for creating multiple output styles:
XML output consists of a hierarchical set of elements, each encoded with a start tag and an end tag. The element should be named for data value that it is encoding:
JSON output consists of a hierarchical set of objects and lists, each encoded with a quoted name, a colon, and a value. If the value is a string, it must be quoted, but numbers are not quoted. Objects are encoded using braces; lists are encoded using square brackets. Data inside objects and lists is separated using commas:
HTML output is designed to allow the output to be rendered in a web browser with minimal effort. Each piece of output data is rendered inside a <div> element, with a class name related to the role of the data. By using a small set of class attribute values, a CSS stylesheet can render the HTML into rich text that mirrors the traditional text content.
Additional attributes can be enabled to provide more details about the data, including data type, description, and an XPath location.
libxo uses format strings to control the rendering of data into the various output styles. Each format string contains a set of zero or more field descriptions, which describe independent data fields. Each field description contains a set of modifiers, a content string, and zero, one, or two format descriptors. The modifiers tell libxo what the field is and how to treat it, while the format descriptors are formatting instructions using printf-style format strings, telling libxo how to format the field. The field description is placed inside a set of braces, with a colon (":") after the modifiers and a slash ("/") before each format descriptors. Text may be intermixed with field descriptions within the format string.
The role describes the function of the field, while the modifiers enable optional behaviors. The contents, field-format, and encoding-format are used in varying ways, based on the role. These are described in the following sections.
In the following example, three field descriptors appear. The first is a padding field containing three spaces of padding, the second is a label ("In stock"), and the third is a value field ("in‑stock"). The in-stock field has a "%u" format that will parse the next argument passed to the xo_emit function as an unsigned integer.
This single line of code can generate text (" In stock: 65\n"), XML ("<in‑stock>65</in‑stock>"), JSON ('"in‑stock": 6'), or HTML (too lengthy to be listed here).
While roles and modifiers typically use single character for brevity, there are alternative names for each which allow more verbose formatting strings. These names must be preceded by a comma, and may follow any single-character values:
The color content can be either static, when placed directly within the field descriptor, or a printf-style format descriptor can be used, if preceded by a slash ("/"):
When using colors, the developer should remember that users will change the foreground and background colors of terminal session according to their own tastes, so assuming that "blue" looks nice is never safe, and is a constant annoyance to your dear author. In addition, a significant percentage of users (1 in 12) will be color blind. Depending on color to convey critical information is not a good idea. Color should enhance output, but should not be used as the sole means of encoding information.
Decorations are typically punctuation marks such as colons, semi-colons, and commas used to decorate the text and make it simpler for human readers. By marking these distinctly, HTML usage scenarios can use CSS to direct their display parameters.
libxo supports internationalization (i18n) through its use of gettext(3). Use the "{G:}" role to request that the remaining part of the format string, following the "{G:}" field, be handled using gettext().
Since gettext() uses the string as the key into the message catalog, libxo uses a simplified version of the format string that removes unimportant field formatting and modifiers, stopping minor formatting changes from impacting the expensive translation process. A developer change such as changing "/%06d" to "/%08d" should not force hand inspection of all .po files.
The simplified version can be generated for a single message using the "xopo -s <text>" command, or an entire .pot can be translated using the "xopo -f <input> -o <output>" command.
xo_emit("{G:}Invalid token\n");
The {G:} role allows a domain name to be set. gettext calls will continue to use that domain name until the current format string processing is complete, enabling a library function to emit strings using it's own catalog. The domain name can be either static as the content of the field, or a format can be used to get the domain name from the arguments.
xo_emit("{G:libc}Service unavailable in restricted mode\n");
Padding represents whitespace used before and between fields.
The padding content can be either static, when placed directly within the field descriptor, or a printf-style format descriptor can be used, if preceded by a slash ("/"):
Title are heading or column headers that are meant to be displayed to the user. The title can be either static, when placed directly within the field descriptor, or a printf-style format descriptor can be used, if preceded by a slash ("/"):
Title fields have an extra convenience feature; if both content and format are specified, instead of looking to the argument list for a value, the content is used, allowing a mixture of format and content within the field descriptor:
xo_emit("{T:Name/%20s}{T:Count/%6s}\n");
Since the incoming argument is a string, the format must be "%s" or something suitable.
Units are the dimension by which values are measured, such as degrees, miles, bytes, and decibels. The units field carries this information for the previous value field.
The value role is used to represent the a data value that is interesting for the non-display output styles (XML and JSON). Value is the default role; if no other role designation is given, the field is a value. The field name must appear within the field descriptor, followed by one or two format descriptors. The first format descriptor is used for display styles (TEXT and HTML), while the second one is used for encoding styles (XML and JSON). If no second format is given, the encoding format defaults to the first format, with any minimum width removed. If no first format is given, both format descriptors default to "%s".
xo_emit("{:length/%02u}x{:width/%02u}x{:height/%02u}\n",
length, width, height);
xo_emit("{:author} wrote \"{:poem}\" in {:year/%4d}\n,
author, poem, year);
The anchor roles allow a set of strings by be padded as a group, but still be visible to xo_emit as distinct fields. Either the start or stop anchor can give a field width and it can be either directly in the descriptor or passed as an argument. Any fields between the start and stop anchor are padded to meet the minimum width given.
To give a width directly, encode it as the content of the anchor tag:
To pass a width as an argument, use "%d" as the format, which must appear after the "/". Note that only "%d" is supported for widths. Using any other value could ruin your day.
If the width is negative, padding will be added on the right, suitable for left justification. Otherwise the padding will be added to the left of the fields between the start and stop anchors, suitable for right justification. If the width is zero, nothing happens. If the number of columns of output between the start and stop anchors is less than the absolute value of the given width, nothing happens.
Widths over 8k are considered probable errors and not supported. If XOF_WARN is set, a warning will be generated.
Field modifiers are flags which modify the way content emitted for particular output styles:
M
Name
Description
a
argument
The content appears as a 'const char *' argument
c
colon
A colon (":") is appended after the label
d
display
Only emit field for display styles (text/HTML)
e
encoding
Only emit for encoding styles (XML/JSON)
g
gettext
Call gettext on field's render content
h
humanize (hn)
Format large numbers in human-readable style
hn-space
Humanize: Place space between numeric and unit
hn-decimal
Humanize: Add a decimal digit, if number < 10
hn-1000
Humanize: Use 1000 as divisor instead of 1024
k
key
Field is a key, suitable for XPath predicates
l
leaf-list
Field is a leaf-list
n
no-quotes
Do not quote the field when using JSON style
p
plural
Gettext: Use comma-separated plural form
q
quotes
Quote the field when using JSON style
t
trim
Trim leading and trailing whitespace
w
white
A blank (" ") is appended after the label
Roles and modifiers can also use more verbose names, when preceded by a comma. For example, the modifier string "Lwc" (or "L,white,colon") means the field has a label role (text that describes the next field) and should be followed by a colon ('c') and a space ('w'). The modifier string "Vkq" (or ":key,quote") means the field has a value role (the default role), that it is a key for the current instance, and that the value should be quoted when encoded for JSON.
The argument modifier indicates that the content of the field descriptor will be placed as a UTF-8 string (const char *) argument within the xo_emit parameters.
EXAMPLE:
xo_emit("{La:} {a:}\n", "Label text", "label", "value");
TEXT:
Label text value
JSON:
"label": "value"
XML:
<label>value</label>
The argument modifier allows field names for value fields to be passed on the stack, avoiding the need to build a field descriptor using snprintf. For many field roles, the argument modifier is not needed, since those roles have specific mechanisms for arguments, such as "{C:fg‑%s}".
The colon modifier is only used for the TEXT and HTML output styles. It is commonly combined with the space modifier ('{w:}'). It is purely a convenience feature.
The gettext modifier is used to translate individual fields using the gettext domain (typically set using the "{G:}" role) and current language settings. Once libxo renders the field value, it is passed to gettext(3), where it is used as a key to find the native language translation.
In the following example, the strings "State" and "full" are passed to gettext() to find locale-based translated strings.
The humanize modifier is used to render large numbers as in a human-readable format. While numbers like "44470272" are completely readable to computers and savants, humans will generally find "44M" more meaningful.
"hn" can be used as an alias for "humanize".
The humanize modifier only affects display styles (TEXT and HMTL). The "no‑humanize" option (See Command-line Arguments) will block the function of the humanize modifier.
There are a number of modifiers that affect details of humanization. These are only available in as full names, not single characters. The "hn‑space" modifier places a space between the number and any multiplier symbol, such as "M" or "K" (ex: "44 K"). The "hn‑decimal" modifier will add a decimal point and a single tenths digit when the number is less than 10 (ex: "4.4K"). The "hn‑1000" modifier will use 1000 as divisor instead of 1024, following the JEDEC-standard instead of the more natural binary powers-of-two tradition.
The key modifier is used to indicate that a particular field helps uniquely identify an instance of list data.
EXAMPLE:
xo_open_list("user");
for (i = 0; i < num_users; i++) {
xo_open_instance("user");
xo_emit("User {k:name} has {:count} tickets\n",
user[i].u_name, user[i].u_tickets);
xo_close_instance("user");
}
xo_close_list("user");
Currently the key modifier is only used when generating XPath value for the HTML output style when XOF_XPATH is set, but other uses are likely in the near future.
The leaf-list modifier is used to distinguish lists where each instance consists of only a single value. In XML, these are rendered as single elements, where JSON renders them as arrays.
EXAMPLE:
for (i = 0; i < num_users; i++) {
xo_emit("Member {l:user}\n", user[i].u_name);
}
XML:
<user>phil</user>
<user>pallavi</user>
JSON:
"user": [ "phil", "pallavi" ]
The name of the field must match the name of the leaf list.
The no-quotes modifier (and its twin, the 'quotes' modifier) affect the quoting of values in the JSON output style. JSON uses quotes for string value, but no quotes for numeric, boolean, and null data. xo_emit applies a simple heuristic to determine whether quotes are needed, but often this needs to be controlled by the caller.
The plural modifier selects the appropriate plural form of an expression based on the most recent number emitted and the current language settings. The contents of the field should be the singular and plural English values, separated by a comma:
xo_emit("{:bytes} {Ngp:byte,bytes}\n", bytes);
The plural modifier is meant to work with the gettext modifier ({g:}) but can work independently. See The Gettext Modifier ({g:}).
When used without the gettext modifier or when the message does not appear in the message catalog, the first token is chosen when the last numeric value is equal to 1; otherwise the second value is used, mimicking the simple pluralization rules of English.
When used with the gettext modifier, the ngettext(3) function is called to handle the heavy lifting, using the message catalog to convert the singular and plural forms into the native language.
The quotes modifier (and its twin, the 'no‑quotes' modifier) affect the quoting of values in the JSON output style. JSON uses quotes for string value, but no quotes for numeric, boolean, and null data. xo_emit applies a simple heuristic to determine whether quotes are needed, but often this needs to be controlled by the caller.
The white space modifier appends a single space to the data value:
EXAMPLE:
xo_emit("{Lw:Name}{:name}\n", "phil");
TEXT:
Name phil
The white space modifier is only used for the TEXT and HTML output styles. It is commonly combined with the colon modifier ('{c:}'). It is purely a convenience feature.
Note that the sense of the 'w' modifier is reversed for the units role ({Uw:}); a blank is added before the contents, rather than after it.
The field format is similar to the format string for printf(3). Its use varies based on the role of the field, but generally is used to format the field's contents.
If the format string is not provided for a value field, it defaults to "%s".
Note a field definition can contain zero or more printf-style 'directives', which are sequences that start with a '%' and end with one of following characters: "diouxXDOUeEfFgGaAcCsSp". Each directive is matched by one of more arguments to the xo_emit function.
The format string has the form:
'%' format-modifier * format-character
The format- modifier can be:
a '#' character, indicating the output value should be prefixed with '0x', typically to indicate a base 16 (hex) value.
a minus sign ('‑'), indicating the output value should be padded on the right instead of the left.
a leading zero ('0') indicating the output value should be padded on the left with zeroes instead of spaces (' ').
one or more digits ('0' - '9') indicating the minimum width of the argument. If the width in columns of the output value is less than the minimum width, the value will be padded to reach the minimum.
a period followed by one or more digits indicating the maximum number of bytes which will be examined for a string argument, or the maximum width for a non-string argument. When handling ASCII strings this functions as the field width but for multi-byte characters, a single character may be composed of multiple bytes. xo_emit will never dereference memory beyond the given number of bytes.
a second period followed by one or more digits indicating the maximum width for a string argument. This modifier cannot be given for non-string arguments.
one or more 'h' characters, indicating shorter input data.
one or more 'l' characters, indicating longer input data.
a 'z' character, indicating a 'size_t' argument.
a 't' character, indicating a 'ptrdiff_t' argument.
a ' ' character, indicating a space should be emitted before positive numbers.
a '+' character, indicating sign should emitted before any number.
Note that 'q', 'D', 'O', and 'U' are considered deprecated and will be removed eventually.
The format character is described in the following table:
Ltr
Argument Type
Format
d
int
base 10 (decimal)
i
int
base 10 (decimal)
o
int
base 8 (octal)
u
unsigned
base 10 (decimal)
x
unsigned
base 16 (hex)
X
unsigned long
base 16 (hex)
D
long
base 10 (decimal)
O
unsigned long
base 8 (octal)
U
unsigned long
base 10 (decimal)
e
double
[-]d.ddde+-dd
E
double
[-]d.dddE+-dd
f
double
[-]ddd.ddd
F
double
[-]ddd.ddd
g
double
as 'e' or 'f'
G
double
as 'E' or 'F'
a
double
[-]0xh.hhhp[+-]d
A
double
[-]0Xh.hhhp[+-]d
c
unsigned char
a character
C
wint_t
a character
s
char *
a UTF-8 string
S
wchar_t *
a unicode/WCS string
p
void *
'%#lx'
The 'h' and 'l' modifiers affect the size and treatment of the argument:
For strings, the 'h' and 'l' modifiers affect the interpretation of the bytes pointed to argument. The default '%s' string is a 'char *' pointer to a string encoded as UTF-8. Since UTF-8 is compatible with ASCII data, a normal 7-bit ASCII string can be used. '%ls' expects a 'wchar_t *' pointer to a wide-character string, encoded as a 32-bit Unicode values. '%hs' expects a 'char *' pointer to a multi-byte string encoded with the current locale, as given by the LC_CTYPE, LANG, or LC_ALL environment varibles. The first of this list of variables is used and if none of the variables are set, the locale defaults to "UTF‑8".
libxo will convert these arguments as needed to either UTF-8 (for XML, JSON, and HTML styles) or locale-based strings for display in text style.
xo_emit("All strings are utf-8 content {:tag/%ls}",
L"except for wide strings");
"%S" is equivalent to "%ls".
Format
Argument Type
Argument Contents
%s
const char *
UTF-8 string
%S
const char *
UTF-8 string (alias for '%s')
%ls
const wchar_t *
Wide character UNICODE string
%hs
const char *
locale-based string
For example, a function is passed a locale-base name, a hat size, and a time value. The hat size is formatted in a UTF-8 (ASCII) string, and the time value is formatted into a wchar_t string.
void print_order (const char *name, int size,
struct tm *timep) {
char buf[32];
const char *size_val = "unknown";
if (size > 0)
snprintf(buf, sizeof(buf), "%d", size);
size_val = buf;
}
wchar_t when[32];
wcsftime(when, sizeof(when), L"%d%b%y", timep);
xo_emit("The hat for {:name/%hs} is {:size/%s}.\n",
name, size_val);
xo_emit("It was ordered on {:order-time/%ls}.\n",
when);
}
It is important to note that xo_emit will perform the conversion required to make appropriate output. Text style output uses the current locale (as described above), while XML, JSON, and HTML use UTF-8.
UTF-8 and locale-encoded strings can use multiple bytes to encode one column of data. The traditional "precision'" (aka "max‑width") value for "%s" printf formatting becomes overloaded since it specifies both the number of bytes that can be safely referenced and the maximum number of columns to emit. xo_emit uses the precision as the former, and adds a third value for specifying the maximum number of columns.
In this example, the name field is printed with a minimum of 3 columns and a maximum of 6. Up to ten bytes of data at the location given by 'name' are in used in filling those columns.
Characters in the format string that are not part of a field definition are copied to the output for the TEXT style, and are ignored for the JSON and XML styles. For HTML, these characters are placed in a <div> with class "text".
EXAMPLE:
xo_emit("The hat is {:size/%s}.\n", size_val);
TEXT:
The hat is extra small.
XML:
<size>extra small</size>
JSON:
"size": "extra small"
HTML:
<div class="text">The hat is </div>
<div class="data" data-tag="size">extra small</div>
<div class="text">.</div>
libxo supports the '%m' directive, which formats the error message associated with the current value of "errno". It is the equivalent of "%s" with the argument strerror(errno).
xo_emit("{:filename} cannot be opened: {:error/%m}", filename);
xo_emit("{:filename} cannot be opened: {:error/%s}",
filename, strerror(errno));
The "eformat" string is the format string used when encoding the field for JSON and XML. If not provided, it defaults to the primary format with any minimum width removed. If the primary is not given, both default to "%s".
Many compilers and tool chains support validation of printf-like arguments. When the format string fails to match the argument list, a warning is generated. This is a valuable feature and while the formatting strings for libxo differ considerably from printf, many of these checks can still provide build-time protection against bugs.
libxo provide variants of functions that provide this ability, if the "‑‑enable‑printflike" option is passed to the "configure" script. These functions use the "_p" suffix, like "xo_emit_p()", xo_emit_hp()", etc.
The following are features of libxo formatting strings that are incompatible with printf-like testing:
implicit formats, where "{:tag}" has an implicit "%s";
the "max" parameter for strings, where "{:tag/%4.10.6s}" means up to ten bytes of data can be inspected to fill a minimum of 4 columns and a maximum of 6;
percent signs in strings, where "{:filled}%" makes a single, trailing percent sign;
the "l" and "h" modifiers for strings, where "{:tag/%hs}" means locale-based string and "{:tag/%ls}" means a wide character string;
distinct encoding formats, where "{:tag/#%s/%s}" means the display styles (text and HTML) will use "#%s" where other styles use "%s";
If none of these features are in use by your code, then using the "_p" variants might be wise.
libxo can retain the parsed internal information related to the given format string, allowing subsequent xo_emit calls, the retained information is used, avoiding repetitive parsing of the format string.
To retain parsed format information, use the XOEF_RETAIN flag to the xo_emit_f() function. A complete set of xo_emit_f functions exist to match all the xo_emit function signatures (with handles, varadic argument, and printf-like flags):
Function
Flags Equivalent
xo_emit_hv
xo_emit_hvf
xo_emit_h
xo_emit_hf
xo_emit
xo_emit_f
xo_emit_hvp
xo_emit_hvfp
xo_emit_hp
xo_emit_hfp
xo_emit_p
xo_emit_fp
The format string must be immutable across multiple calls to xo_emit_f(), since the library retains the string. Typically this is done by using static constant strings, such as string literals. If the string is not immutable, the XOEF_RETAIN flag must not be used.
The functions xo_retain_clear() and xo_retain_clear_all() release internal information on either a single format string or all format strings, respectively. Neither is required, but the library will retain this information until it is cleared or the process exits.
const char *fmt = "{:name} {:count/%d}\n";
for (i = 0; i < 1000; i++) {
xo_open_instance("item");
xo_emit_f(XOEF_RETAIN, fmt, name[i], count[i]);
}
xo_retain_clear(fmt);
The retained information is kept as thread-specific data.
For XML and JSON, individual fields appear inside hierarchies which provide context and meaning to the fields. Unfortunately, these encoding have a basic disconnect between how lists is similar objects are represented.
JSON encodes lists using a single name and square brackets:
"user": [ "phil", "pallavi", "sjg" ]
This means libxo needs three distinct indications of hierarchy: one for containers of hierarchy appear only once for any specific parent, one for lists, and one for each item in a list.
A "container" is an element of a hierarchy that appears only once under any specific parent. The container has no value, but serves to contain other nodes.
To open a container, call xo_open_container() or xo_open_container_h(). The former uses the default handle and the latter accepts a specific handle.
int xo_open_container_h (xo_handle_t *xop, const char *name);
int xo_open_container (const char *name);
To close a level, use the xo_close_container() or xo_close_container_h() functions:
int xo_close_container_h (xo_handle_t *xop, const char *name);
int xo_close_container (const char *name);
Each open call must have a matching close call. If the XOF_WARN flag is set and the name given does not match the name of the currently open container, a warning will be generated.
A list is set of one or more instances that appear under the same parent. The instances contain details about a specific object. One can think of instances as objects or records. A call is needed to open and close the list, while a distinct call is needed to open and close each instance of the list:
Some users may find tracking the names of open containers, lists, and instances inconvenient. libxo offers a "Do The Right Thing" mode, where libxo will track the names of open containers, lists, and instances so the close function can be called without a name. To enable DTRT mode, turn on the XOF_DTRT flag prior to making any other libxo output.
xo_set_flags(NULL, XOF_DTRT);
Each open and close function has a version with the suffix "_d", which will close the open container, list, or instance:
xo_open_list("item");
for (...) {
xo_open_instance("item");
xo_emit(...);
xo_close_instance_d();
}
xo_close_list_d();
Note that the XOF_WARN flag will also cause libxo to track open containers, lists, and instances. A warning is generated when the name given to the close function and the name recorded do not match.
Markers are used to protect and restore the state of open constructs. While a marker is open, no other open constructs can be closed. When a marker is closed, all constructs open since the marker was opened will be closed.
Markers use names which are not user-visible, allowing the caller to choose appropriate internal names.
In this example, the code whiffles through a list of fish, calling a function to emit details about each fish. The marker "fish‑guts" is used to ensure that any constructs opened by the function are closed properly.
for (i = 0; fish[i]; i++) {
xo_open_instance("fish");
xo_open_marker("fish-guts");
dump_fish_details(i);
xo_close_marker("fish-guts");
}
The "colors" option takes a value that is a set of mappings from the pre-defined set of colors to new foreground and background colors. The value is a series of "fg/bg" values, separated by a "+". Each pair of "fg/bg" values gives the colors to which a basic color is mapped when used as a foreground or background color. The order is the mappings is:
black
red
green
yellow
blue
magenta
cyan
white
Pairs may be skipped, leaving them mapped as normal, as are missing pairs or single colors.
For example consider the following xo_emit call:
xo_emit("{C:fg-red,bg-green}Merry XMas!!{C:}\n");
To turn all colored output to red-on-blue, use eight pairs of "red/blue" mappings separated by "+"s:
To turn the red-on-green text to magenta-on-cyan, give a "magenta" foreground value for red (the second mapping) and a "cyan" background to green (the third mapping):
--libxo colors=+magenta+/cyan
Consider the common situation where blue output looks unreadable on a terminal session with a black background. To turn both "blue" foreground and background output to "yellow", give only the fifth mapping, skipping the first four mappings with bare "+"s:
libxo uses "handles" to control its rendering functionality. The handle contains state and buffered data, as well as callback functions to process data.
Handles give an abstraction for libxo that encapsulates the state of a stream of output. Handles have the data type "xo_handle_t" and are opaque to the caller.
The library has a default handle that is automatically initialized. By default, this handle will send text style output (XO_STYLE_TEXT) to standard output. The xo_set_style and xo_set_flags functions can be used to change this behavior.
For the typical command that is generating output on standard output, there is no need to create an explicit handle, but they are available when needed, e.g., for daemons that generate multiple streams of output.
Many libxo functions take a handle as their first parameter; most that do not use the default handle. Any function taking a handle can be passed NULL to access the default handle. For the convenience of callers, the libxo library includes handle-less functions that implicitly use the default handle.
For example, the following are equivalent:
xo_emit("test");
xo_emit_h(NULL, "test");
Handles are created using xo_create() and destroy using xo_destroy().
By default, libxo writes output to standard output. A convenience function is provided for situations when output should be written to a different file:
The xo_set_writer function allows custom 'write' functions which can tailor how libxo writes data. An opaque argument is recorded and passed back to the write function, allowing the function to acquire context information. The 'close' function can release this opaque data and any other resources as needed. The flush function can flush buffered data associated with the opaque object.
The XOF_CLOSE_FP flag will trigger the call of the close_func (provided via xo_set_writer()) when the handle is destroyed.
The XOF_COLOR flag enables color and effects in output regardless of output device, while the XOF_COLOR_ALLOWED flag allows color and effects only if the output device is a terminal.
The XOF_PRETTY flag requests 'pretty printing', which will trigger the addition of indentation and newlines to enhance the readability of XML, JSON, and HTML output. Text output is not affected.
The XOF_WARN flag requests that warnings will trigger diagnostic output (on standard error) when the library notices errors during operations, or with arguments to functions. Without warnings enabled, such conditions are ignored.
Warnings allow developers to debug their interaction with libxo. The function "xo_failure" can used as a breakpoint for a debugger, regardless of whether warnings are enabled.
If the style is XO_STYLE_HTML, the following additional flags can be used:
Flag
Description
XOF_XPATH
Emit "data-xpath" attributes
XOF_INFO
Emit additional info fields
The XOF_XPATH flag enables the emission of XPath expressions detailing the hierarchy of XML elements used to encode the data field, if the XPATH style of output were requested.
The XOF_INFO flag encodes additional informational fields for HTML output. See Field Information (xo_info_t) for details.
If the style is XO_STYLE_XML, the following additional flags can be used:
Flag
Description
XOF_KEYS
Flag 'key' fields for xml
The XOF_KEYS flag adds 'key' attribute to the XML encoding for field definitions that use the 'k' modifier. The key attribute has the value "key":
The xo_destroy function releases a handle and any resources it is using. Calling xo_destroy with a NULL handle will release any resources associated with the default handle.
int xo_emit (const char *fmt, ...);
int xo_emit_h (xo_handle_t *xop, const char *fmt, ...);
int xo_emit_hv (xo_handle_t *xop, const char *fmt, va_list vap);
The "fmt" argument is a string containing field descriptors as specified in Format Strings. The use of a handle is optional and NULL can be passed to access the internal 'default' handle. See Handles.
The remaining arguments to xo_emit() and xo_emit_h() are a set of arguments corresponding to the fields in the format string. Care must be taken to ensure the argument types match the fields in the format string, since an inappropriate cast can ruin your day. The vap argument to xo_emit_hv() points to a variable argument list that can be used to retrieve arguments via va_arg().
These functions are intended to avoid the scenario where one would otherwise need to compose a format descriptors using snprintf(). The individual parts of the format descriptor are passed in distinctly.
xo_emit("T", "Host name is ", NULL, NULL);
xo_emit("V", "host-name", NULL, NULL, host-name);
The name parameter give the name of the attribute to be encoded. The fmt parameter gives a printf-style format string used to format the value of the attribute using any remaining arguments, or the vap parameter passed to xo_attr_hv().
xo_attr is placed on the next container, instance, leaf, or leaf list that is emitted.
Since attributes are only emitted in XML, their use should be limited to meta-data and additional or redundant representations of data already emitted in other form.
libxo buffers data, both for performance and consistency, but also to allow some advanced features to work properly. At various times, the caller may wish to flush any data buffered within the library. The xo_flush() call is used for this:
When the program is ready to exit or close a handle, a call to xo_finish() is required. This flushes any buffered data, closes open libxo constructs, and completes any pending operations.
int xo_finish (void);
int xo_finish_h (xo_handle_t *xop);
void xo_finish_atexit (void);
Calling this function is vital to the proper operation of libxo, especially for the non-TEXT output styles.
xo_finish_atexit is suitable for use with atexit(3).
libxo represents to types of hierarchy: containers and lists. A container appears once under a given parent where a list contains instances that can appear multiple times. A container is used to hold related fields and to give the data organization and scope.
To create a container, use the xo_open_container and xo_close_container functions:
int xo_open_container (const char *name);
int xo_open_container_h (xo_handle_t *xop, const char *name);
int xo_open_container_hd (xo_handle_t *xop, const char *name);
int xo_open_container_d (const char *name);
int xo_close_container (const char *name);
int xo_close_container_h (xo_handle_t *xop, const char *name);
int xo_close_container_hd (xo_handle_t *xop);
int xo_close_container_d (void);
The name parameter gives the name of the container, encoded in UTF-8. Since ASCII is a proper subset of UTF-8, traditional C strings can be used directly.
The close functions with the "_d" suffix are used in "Do The Right Thing" mode, where the name of the open containers, lists, and instances are maintained internally by libxo to allow the caller to avoid keeping track of the open container name.
Use the XOF_WARN flag to generate a warning if the name given on the close does not match the current open container.
For TEXT and HTML output, containers are not rendered into output text, though for HTML they are used when the XOF_XPATH flag is set.
EXAMPLE:
xo_open_container("system");
xo_emit("The host name is {:host-name}\n", hn);
xo_close_container("system");
XML:
<system><host-name>foo</host-name></system>
Lists are sequences of instances of homogeneous data objects. Two distinct levels of calls are needed to represent them in our output styles. Calls must be made to open and close a list, and for each instance of data in that list, calls must be make to open and close that instance.
The name given to all calls must be identical, and it is strongly suggested that the name be singular, not plural, as a matter of style and usage expectations.
The xo_parse_args() function is used to process a program's arguments. libxo-specific options are processed and removed from the argument list so the calling application does not need to process them. If successful, a new value for argc is returned. On failure, a message it emitted and -1 is returned.
argc = xo_parse_args(argc, argv);
if (argc < 0)
exit(EXIT_FAILURE);
Following the call to xo_parse_args, the application can process the remaining arguments in a normal manner. See Command-line Arguments for a description of valid arguments.
The xo_set_program function sets name of the program as reported by functions like xo_failure, xo_warn, xo_err, etc. The program name is initialized by xo_parse_args, but subsequent calls to xo_set_program can override this value.
xo_set_program(argv[0]);
Note that the value is not copied, so the memory passed to xo_set_program (and xo_parse_args) must be maintained by the caller.
The xo_set_version function records a version number to be emitted as part of the data for encoding styles (XML and JSON). This version number is suitable for tracking changes in the content, allowing a user of the data to discern which version of the data model is in use.
HTML data can include additional information in attributes that begin with "data‑". To enable this, three things must occur:
First the application must build an array of xo_info_t structures, one per tag. The array must be sorted by name, since libxo uses a binary search to find the entry that matches names from format instructions.
Second, the application must inform libxo about this information using the xo_set_info() call:
typedef struct xo_info_s {
const char *xi_name; /* Name of the element */
const char *xi_type; /* Type of field */
const char *xi_help; /* Description of field */
} xo_info_t;
void xo_set_info (xo_handle_t *xop, xo_info_t *infop, int count);
Like other libxo calls, passing NULL for the handle tells libxo to use the default handle.
If the count is -1, libxo will count the elements of infop, but there must be an empty element at the end. More typically, the number is known to the application:
xo_info_t info[] = {
{ "in-stock", "number", "Number of items in stock" },
{ "name", "string", "Name of the item" },
{ "on-order", "number", "Number of items on order" },
{ "sku", "string", "Stock Keeping Unit" },
{ "sold", "number", "Number of items sold" },
};
int info_count = (sizeof(info) / sizeof(info[0]));
...
xo_set_info(NULL, info, info_count);
Third, the emission of info must be triggered with the XOF_INFO flag using either the xo_set_flags() function or the "‑‑libxo=info" command line argument.
The type and help values, if present, are emitted as the "data‑type" and "data‑help" attributes:
realloc_func should expect the same arguments as realloc(3) and return a pointer to memory following the same convention. free_func will receive the same argument as free(3) and should release it, as appropriate for the environment.
By default, the standard realloc() and free() functions are used.
The environment variable "LIBXO_OPTIONS" can be set to a subset of libxo options, including:
color
flush
flush-line
no-color
no-humanize
no-locale
no-retain
pretty
retain
underscores
warn
For example, warnings can be enabled by:
% env LIBXO_OPTIONS=warn my-app
Since environment variables are inherited, child processes will have the same options, which may be undesirable, making the use of the "‑‑libxo" option is preferable in most situations.
Many programs make use of the standard library functions err() and warn() to generate errors and warnings for the user. libxo wants to pass that information via the current output style, and provides compatible functions to allow this:
These functions display the program name, a colon, a formatted message based on the arguments, and then optionally a colon and an error message associated with either "errno" or the "code" parameter.
EXAMPLE:
if (open(filename, O_RDONLY) < 0)
xo_err(1, "cannot open file '%s'", filename);
The xo_error function can be used for generic errors that should be reported over the handle, rather than to stderr. The xo_error function behaves like xo_err for TEXT and HTML output styles, but puts the error into XML or JSON elements:
EXAMPLE::
xo_error("Does not %s", "compute");
XML::
<error><message>Does not compute</message></error>
JSON::
"error": { "message": "Does not compute" }
libxo automatically initializes the locale based on setting of the environment variables LC_CTYPE, LANG, and LC_ALL. The first of this list of variables is used and if none of the variables, the locale defaults to "UTF‑8". The caller may wish to avoid this behavior, and can do so by calling the xo_no_setlocale() function.
syslog is the system logging facility used throughout the unix world. Messages are sent from commands, applications, and daemons to a hierarchy of servers, where they are filtered, saved, and forwarded based on configuration behaviors.
syslog is an older protocol, originally documented only in source code. By the time RFC 3164 published, variation and mutation left the leading "<pri>" string as only common content. RFC 5424 defines a new version (version 1) of syslog and introduces structured data into the messages. Structured data is a set of name/value pairs transmitted distinctly alongside the traditional text message, allowing filtering on precise values instead of regular expressions.
These name/value pairs are scoped by a two-part identifier; an enterprise identifier names the party responsible for the message catalog and a name identifying that message. Enterprise IDs are defined by IANA, the Internet Assigned Numbers Authority:
The message name should follow the conventions in What makes a good field name?, as should the fields within the message.
/* Both of these calls are optional */
xo_set_syslog_enterprise_id(32473);
xo_open_log("my-program", 0, LOG_DAEMON);
/* Generate a syslog message */
xo_syslog(LOG_ERR, "upload-failed",
"error <%d> uploading file '{:filename}' "
"as '{:target/%s:%s}'",
code, filename, protocol, remote);
xo_syslog(LOG_INFO, "poofd-invalid-state",
"state {:current/%u} is invalid {:connection/%u}",
state, conn);
The developer should be aware that the message name may be used in the future to allow access to further information, including documentation. Care should be taken to choose quality, descriptive names.
The xo_syslog, xo_vsyslog, and xo_open_log functions accept a set of flags which provide the priority of the message, the source facility, and some additional features. These values are OR'd together to create a single integer argument:
xo_syslog(LOG_ERR | LOG_AUTH, "login-failed",
"Login failed; user '{:user}' from host '{:address}'",
user, addr);
These values are defined in <syslog.h>.
The priority value indicates the importance and potential impact of each message.
Priority
Description
LOG_EMERG
A panic condition, normally broadcast to all users
LOG_ALERT
A condition that should be corrected immediately
LOG_CRIT
Critical conditions
LOG_ERR
Generic errors
LOG_WARNING
Warning messages
LOG_NOTICE
Non-error conditions that might need special handling
LOG_INFO
Informational messages
LOG_DEBUG
Developer-oriented messages
The facility value indicates the source of message, in fairly generic terms.
Facility
Description
LOG_AUTH
The authorization system (e.g. login(1))
LOG_AUTHPRIV
As LOG_AUTH, but logged to a privileged file
LOG_CRON
The cron daemon: cron(8)
LOG_DAEMON
System daemons, not otherwise explicitly listed
LOG_FTP
The file transfer protocol daemons
LOG_KERN
Messages generated by the kernel
LOG_LPR
The line printer spooling system
LOG_MAIL
The mail system
LOG_NEWS
The network news system
LOG_SECURITY
Security subsystems, such as ipfw(4)
LOG_SYSLOG
Messages generated internally by syslogd(8)
LOG_USER
Messages generated by user processes (default)
LOG_UUCP
The uucp system
LOG_LOCAL0..7
Reserved for local use
In addition to the values listed above, xo_open_log accepts a set of addition flags requesting specific behaviors.
Flag
Description
LOG_CONS
If syslogd fails, attempt to write to /dev/console
Use the xo_syslog function to generate syslog messages by calling it with a log priority and facility, a message name, a format string, and a set of arguments. The priority/facility argument are discussed above, as is the message name.
The format string follows the same conventions as xo_emit's format string, with each field being rendered as an SD-PARAM pair.
xo_syslog(LOG_ERR, "poofd-missing-file",
"'{:filename}' not found: {:error/%m}", filename);
... [poofd-missing-file@32473 filename="/etc/poofd.conf"
error="Permission denied"] '/etc/poofd.conf' not
found: Permission denied
xo_open_log functions similar to openlog(3), allowing customization of the program name, the log facility number, and the additional option flags described in Priority, Facility, and Flags.
void
xo_open_log (const char *ident, int logopt, int facility);
xo_set_logmask function similar to setlogmask(3), restricting the set of generated log event to those whose associated bit is set in maskpri. Use LOG_MASK(pri) to find the appropriate bit, or LOG_UPTO(toppri) to create a mask for all priorities up to and including toppri.
int
xo_set_logmask (int maskpri);
Example:
setlogmask(LOG_UPTO(LOG_WARN));
Use the xo_set_syslog_enterprise_id to supply a platform- or application-specific enterprise id. This value is used in any future syslog messages.
Ideally, the operating system should supply a default value via the "kern.syslog.enterprise_id" sysctl value. Lacking that, the application should provide a suitable value.
void
xo_set_syslog_enterprise_id (unsigned short eid);
Enterprise IDs are administered by IANA, the Internet Assigned Number Authority. The complete list is EIDs on their web site:
Each software development organization that defines a set of syslog messages should register their own EID and use that value in their software to ensure that messages can be uniquely identified by the combination of EID + message name.
The number of encoding schemes in current use is staggering, with new and distinct schemes appearing daily. While libxo provide XML, JSON, HMTL, and text natively, there are requirements for other encodings.
Rather than bake support for all possible encoders into libxo, the API allows them to be defined externally. libxo can then interfaces with these encoding modules using a simplistic API. libxo processes all functions calls, handles state transitions, performs all formatting, and then passes the results as operations to a customized encoding function, which implements specific encoding logic as required. This means your encoder doesn't need to detect errors with unbalanced open/close operations but can rely on libxo to pass correct data.
By making a simple API, libxo internals are not exposed, insulating the encoder and the library from future or internal changes.
The three elements of the API are:
loading
initialization
operations
The following sections provide details about these topics.
libxo source contain an encoder for Concise Binary Object Representation, aka CBOR (RFC 7049) which can be used as used as an example for the API.
Encoders can be registered statically or discovered dynamically. Applications can choose to call the xo_encoder_register() function to explicitly register encoders, but more typically they are built as shared libraries, placed in the libxo/extensions directory, and loaded based on name. libxo looks for a file with the name of the encoder and an extension of ".enc". This can be a file or a symlink to the shared library file that supports the encoder.
% ls -1 lib/libxo/extensions/*.enc
lib/libxo/extensions/cbor.enc
lib/libxo/extensions/test.enc
Each encoder must export a symbol used to access the library, which must have the following signature:
int xo_encoder_library_init (XO_ENCODER_INIT_ARGS);
XO_ENCODER_INIT_ARGS is a macro defined in xo_encoder.h that defines an argument called "arg", a pointer of the type xo_encoder_init_args_t. This structure contains two fields:
xei_version is the version number of the API as implemented within libxo. This version is currently as 1 using XO_ENCODER_VERSION. This number can be checked to ensure compatibility. The working assumption is that all versions should be backward compatible, but each side may need to accurately know the version supported by the other side. xo_encoder_library_init can optionally check this value, and must then set it to the version number used by the encoder, allowing libxo to detect version differences and react accordingly. For example, if version 2 adds new operations, then libxo will know that an encoding library that set xei_version to 1 cannot be expected to handle those new operations.
xei_handler must be set to a pointer to a function of type xo_encoder_func_t, as defined in xo_encoder.h. This function takes a set of parameters: -- xop is a pointer to the opaque xo_handle_t structure -- op is an integer representing the current operation -- name is a string whose meaning differs by operation -- value is a string whose meaning differs by operation -- private is an opaque structure provided by the encoder
Additional arguments may be added in the future, so handler functions should use the XO_ENCODER_HANDLER_ARGS macro. An appropriate "extern" declaration is provided to help catch errors.
Once the encoder initialization function has completed processing, it should return zero to indicate that no error has occurred. A non-zero return code will cause the handle initialization to fail.
The encoder API defines a set of operations representing the processing model of libxo. Content is formatted within libxo, and callbacks are made to the encoder's handler function when data is ready to be processed.
Operation
Meaning (Base function)
XO_OP_CREATE
Called when the handle is created
XO_OP_OPEN_CONTAINER
Container opened (xo_open_container)
XO_OP_CLOSE_CONTAINER
Container closed (xo_close_container)
XO_OP_OPEN_LIST
List opened (xo_open_list)
XO_OP_CLOSE_LIST
List closed (xo_close_list)
XO_OP_OPEN_LEAF_LIST
Leaf list opened (xo_open_leaf_list)
XO_OP_CLOSE_LEAF_LIST
Leaf list closed (xo_close_leaf_list)
XO_OP_OPEN_INSTANCE
Instance opened (xo_open_instance)
XO_OP_CLOSE_INSTANCE
Instance closed (xo_close_instance)
XO_OP_STRING
Field with Quoted UTF-8 string
XO_OP_CONTENT
Field with content
XO_OP_FINISH
Finish any pending output
XO_OP_FLUSH
Flush any buffered output
XO_OP_DESTROY
Clean up resources
XO_OP_ATTRIBUTE
An attribute name/value pair
XO_OP_VERSION
A version string
For all the open and close operations, the name parameter holds the name of the construct. For string, content, and attribute operations, the name parameter is the name of the field and the value parameter is the value. "string" are differentiated from "content" to allow differing treatment of true, false, null, and numbers from real strings, though content values are formatted as strings before the handler is called. For version operations, the value parameter contains the version.
The "xo" utility allows command line access to the functionality of the libxo library. Using "xo", shell scripts can emit XML, JSON, and HTML using the same commands that emit text output.
The style of output can be selected using a specific option: "‑X" for XML, "‑J" for JSON, "‑H" for HTML, or "‑T" for TEXT, which is the default. The "--style <style>" option can also be used. The standard set of "‑‑libxo" options are available (see Command-line Arguments), as well as the LIBXO_OPTIONS environment variable (see LIBXO_OPTIONS).
The "xo" utility accepts a format string suitable for xo_emit() and a set of zero or more arguments used to supply data for that string.
xo "The {k:name} weighs {:weight/%d} pounds.\n" fish 6
TEXT:
The fish weighs 6 pounds.
XML:
<name>fish</name>
<weight>6</weight>
JSON:
"name": "fish",
"weight": 6
HTML:
<div class="line">
<div class="text">The </div>
<div class="data" data-tag="name">fish</div>
<div class="text"> weighs </div>
<div class="data" data-tag="weight">6</div>
<div class="text"> pounds.</div>
</div>
The "--wrap <path>" option can be used to wrap emitted content in a specific hierarchy. The path is a set of hierarchical names separated by the '/' character.
The "--open <path>" and "--close <path>" can be used to emit hierarchical information without the matching close and open tag. This allows a shell script to emit open tags, data, and then close tags. The "‑‑depth" option may be used to set the depth for indentation. The "‑‑leading‑xpath" may be used to prepend data to the XPath values used for HTML output style.
#!/bin/sh
xo --open top/data
xo --depth 2 '{tag}' value
xo --close top/data
XML:
<top>
<data>
<tag>value</tag>
</data>
</top>
JSON:
"top": {
"data": {
"tag": "value"
}
}
--close <path> Close tags for the given path
--depth <num> Set the depth for pretty printing
--help Display this help text
--html OR -H Generate HTML output
--json OR -J Generate JSON output
--leading-xpath <path> Add a prefix to generated XPaths (HTML)
--open <path> Open tags for the given path
--pretty OR -p Make 'pretty' output (add indent, newlines)
--style <style> Generate given style (xml, json, text, html)
--text OR -T Generate text output (the default style)
--version Display version information
--warn OR -W Display warnings in text on stderr
--warn-xml Display warnings in xml on stdout
--wrap <path> Wrap output in a set of containers
--xml OR -X Generate XML output
--xpath Add XPath data to HTML output);
% xo 'The {:product} is {:status}\n' stereo "in route"
The stereo is in route
% ./xo/xo -p -X 'The {:product} is {:status}\n' stereo "in route"
<product>stereo</product>
<status>in route</status>
xolint is a tool for reporting common mistakes in format strings in source code that invokes xo_emit(). It allows these errors to be diagnosed at build time, rather than waiting until runtime.
xolint takes the one or more C files as arguments, and reports and errors, warning, or informational messages as needed.
Option
Meaning
-c
Invoke 'cpp' against the input file
-C <flags>
Flags that are passed to 'cpp
-d
Enable debug output
-D
Generate documentation for all xolint messages
-I
Generate info table code
-p
Print the offending lines after the message
-V
Print vocabulary of all field names
-X
Extract samples from xolint, suitable for testing
The output message will contain the source filename and line number, the class of the message, the message, and, if -p is given, the line that contains the error:
% xolint.pl -t xolint.c
xolint.c: 16: error: anchor format should be "%d"
16 xo_emit("{[:/%s}");
The "‑I" option will generate a table of xo_info_t structures ,
The "‑V" option does not report errors, but prints a complete list of all field names, sorted alphabetically. The output can help spot inconsistencies and spelling errors.
xohtml is a tool for turning the output of libxo-enabled commands into html files suitable for display in modern HTML web browsers. It can be used to test and debug HTML output, as well as to make the user ache to escape the world of 70s terminal devices.
xohtml is given a command, either on the command line or via the "‑c" option. If not command is given, standard input is used. The command's output is wrapped in HTML tags, with references to supporting CSS and Javascript files, and written to standard output or the file given in the "‑f" option. The "‑b" option can be used to provide an alternative base path for the support files.
Option
Meaning
-b <base>
Base path for finding css/javascript files
-c <command>
Command to execute
-f <file>
Output file name
The "‑c" option takes a full command with arguments, including any libxo options needed to generate html ("‑‑libxo=html"). This value must be quoted if it consists of multiple tokens.
The "xopo" utility filters ".pot" files generated by the "xgettext" utility to remove formatting information suitable for use with the "{G:}" modifier. This means that when the developer changes the formatting portion of the field definitions, or the fields modifiers, the string passed to gettext(3) is unchanged, avoiding the expense of updating any existing translation files (".po" files).
The syntax for the xopo command is one of two forms; it can be used as a filter for processing a .po or .pot file, rewriting the "msgid" strings with a simplified message string. In this mode, the input is either standard input or a file given by the "‑f" option, and the output is either standard output or a file given by the "‑o" option.
In the second mode, a simple message given using the "‑s" option on the command, and the simplified version of that message is printed on stdout.
In 2001, we added an XML API to the JUNOS operating system, which is built on top of FreeBSD. Eventually this API became standardized as the NETCONF API (RFC 6241). As part of this effort, we modified many FreeBSD utilities to emit XML, typically via a "‑X" switch. The results were mixed. The cost of maintaining this code, updating it, and carrying it were non-trivial, and contributed to our expense (and the associated delay) with upgrading the version of FreeBSD on which each release of JUNOS is based.
A recent (2014) effort within JUNOS aims at removing our modifications to the underlying FreeBSD code as a means of reducing the expense and delay in tracking HEAD. JUNOS is structured to have system components generate XML that is rendered by the CLI (think: login shell) into human-readable text. This allows the API to use the same plumbing as the CLI, and ensures that all components emit XML, and that it is emitted with knowledge of the consumer of that XML, yielding an API that have no incremental cost or feature delay.
libxo is an effort to mix the best aspects of the JUNOS strategy into FreeBSD in a seemless way, allowing commands to make printf-like output calls with a single code path.
The history is both long and short: libxo's functionality is based on what JUNOS does in a data modeling language called ODL (output definition language). In JUNOS, all subcomponents generate XML, which is feed to the CLI, where data from the ODL files tell is how to render that XML into text. ODL might had a set of tags like:
tag docsis-state {
help "State of the DOCSIS interface";
type string;
}
tag docsis-mode {
help "DOCSIS mode (2.0/3.0) of the DOCSIS interface";
type string;
}
tag docsis-upstream-speed {
help "Operational upstream speed of the interface";
type string;
}
tag downstream-scanning {
help "Result of scanning in downstream direction";
type string;
}
tag ranging {
help "Result of ranging action";
type string;
}
tag signal-to-noise-ratio {
help "Signal to noise ratio for all channels";
type string;
}
tag power {
help "Operational power of the signal on all channels";
type string;
}
format docsis-status-format {
picture "
State : @, Mode: @, Upstream speed: @
Downstream scanning: @, Ranging: @
Signal to noise ratio: @
Power: @
";
line {
field docsis-state;
field docsis-mode;
field docsis-upstream-speed;
field downstream-scanning;
field ranging;
field signal-to-noise-ratio;
field power;
}
}
These tag definitions are compiled into field definitions that are triggered when matching XML elements are seen. ODL also supports other means of defining output.
The roles and modifiers describe these details.
In moving these ideas to bsd, two things had to happen: the formatting had to happen at the source since BSD won't have a JUNOS-like CLI to do the rendering, and we can't depend on external data models like ODL, which was seen as too hard a sell to the BSD community.
The results were that the xo_emit strings are used to encode the roles, modifiers, names, and formats. They are dense and a bit cryptic, but not so unlike printf format strings that developers will be lost.
libxo is a new implementation of these ideas and is distinct from the previous implementation in JUNOS.
To make useful, consistent field names, follow these guidelines:
Use lower case, even for TLAs
Lower case is more civilized. Even TLAs should be lower case to avoid scenarios where the differences between "XPath" and "Xpath" drive your users crazy. Using "xpath" is simpler and better.
Use hyphens, not underscores
Use of hyphens is traditional in XML, and the XOF_UNDERSCORES flag can be used to generate underscores in JSON, if desired. But the raw field name should use hyphens.
Use full words
Don't abbreviate especially when the abbreviation is not obvious or not widely used. Use "data‑size", not "dsz" or "dsize". Use "interface" instead of "ifname", "if‑name", "iface", "if", or "intf".
Use <verb>-<units>
Using the form <verb>-<units> or <verb>-<classifier>-<units> helps in making consistent, useful names, avoiding the situation where one app uses "sent‑packet" and another "packets‑sent" and another "packets‑we‑have‑sent". The <units> can be dropped when it is obvious, as can obvious words in the classification. Use "receive‑after‑window‑packets" instead of "received‑packets‑of‑data‑after‑window".
Find someone else who is expressing similar data and follow their fields and hierarchy. Remember the quote is not "Consistency is the hobgoblin of little minds", but "A foolish consistency is the hobgoblin of little minds".
Use containment as scoping
In the previous example, all the names are prefixed with "proc‑", which is redundant given that they are nested under the process table.
Think about your users
Have empathy for your users, choosing clear and useful fields that contain clear and useful data. You may need to augment the display content with xo_attr() calls (Attributes (xo_attr)) or "{e:}" fields (The Encoding Modifier ({e:})) to make the data useful.
Don't use an arbitrary number postfix
What does "errors2" mean? No one will know. "errors‑after‑restart" would be a better choice. Think of your users, and think of the future. If you make "errors2", the next guy will happily make "errors3" and before you know it, someone will be asking what's the difference between errors37 and errors63.
Be consistent, uniform, unsurprising, and predictable
Think of your field vocabulary as an API. You want it useful, expressive, meaningful, direct, and obvious. You want the client application's programmer to move between without the need to understand a variety of opinions on how fields are named. They should see the system as a single cohesive whole, not a sack of cats.
Field names constitute the means by which client programmers interact with our system. By choosing wise names now, you are making their lives better.
After using "xolint" to find errors in your field descriptors, use "xolint -V" to spell check your field names and to detect different names for the same data. "dropped‑short" and "dropped‑too‑short" are both reasonable names, but using them both will lead users to ask the difference between the two fields. If there is no difference, use only one of the field names. If there is a difference, change the names to make that difference more obvious.
The message "Format cannot be given when content is present (roles: CDLN)" can be caused by code like:
xo_emit("{N:Max/%6.6s}", "Max");
Fields with the C, D, L, or N roles can't have both static literal content ("{L:Label}") and a format ("{L:/%s}"). This error will also occur when the content has a backslash in it, like "{N:Type of I/O}"; backslashes should be escaped, like "{N:Type of I\\/O}". Note the double backslash, one for handling 'C' strings, and one for libxo.
The message "Field has invalid color or effect (role: C)" can be caused by code like:
xo_emit("{C:fg-purple,bold}{:foo}{C:gween}", x);
This code should be replaced with code like:
xo_emit("{C:fg-red,bold}{:foo}{C:fg-green}", x);
The list of colors and effects are limited. The set of colors includes default, black, red, green, yellow, blue, magenta, cyan, and white, which must be prefixed by either "fg‑" or "bg‑". Effects are limited to bold, no-bold, underline, no-underline, inverse, no-inverse, normal, and reset. Values must be separated by commas.
The message "Use hyphens, not underscores, for value field name" can be caused by code like:
xo_emit("{:no_under_scores}", "bad");
This code should be replaced with code like:
xo_emit("{:no-under-scores}", "bad");
Use of hyphens is traditional in XML, and the XOF_UNDERSCORES flag can be used to generate underscores in JSON, if desired. But the raw field name should use hyphens.
The message "Value field name should be lower case" can be caused by code like:
xo_emit("{:WHY-ARE-YOU-SHOUTING}", "NO REASON");
This code should be replaced with code like:
xo_emit("{:why-are-you-shouting}", "no reason");
Lower case is more civilized. Even TLAs should be lower case to avoid scenarios where the differences between "XPath" and "Xpath" drive your users crazy. Lower case rules the seas.
The message "Value field name should be longer than two characters" can be caused by code like:
xo_emit("{:x}", "mumble");
This code should be replaced with code like:
xo_emit("{:something-meaningful}", "mumble");
Field names should be descriptive, and it's hard to be descriptive in less than two characters. Consider your users and try to make something more useful. Note that this error often occurs when the field type is placed after the colon ("{:T/%20s}"), instead of before it ("{T:/20s}").
The message "Value field name contains invalid character" can be caused by code like:
xo_emit("{:cost-in-$$/%u}", 15);
This code should be replaced with code like:
xo_emit("{:cost-in-dollars/%u}", 15);
An invalid character is often a sign of a typo, like "{:]}" instead of "{]:}". Field names are restricted to lower-case characters, digits, and hyphens.
The message "decoration field contains invalid character" can be caused by code like:
xo_emit("{D:not good}");
This code should be replaced with code like:
xo_emit("{D:((}{:good}{D:))}", "yes");
This is minor, but fields should use proper roles. Decoration fields are meant to hold punctuation and other characters used to decorate the content, typically to make it more readable to human readers.
The message "Anchor content should be decimal width" can be caused by code like:
xo_emit("{[:mumble}");
This code should be replaced with code like:
xo_emit("{[:32}");
Anchors need an integer value to specify the width of the set of anchored fields. The value can be positive (for left padding/right justification) or negative (for right padding/left justification) and can appear in either the start or stop anchor field descriptor.
The message "Anchor format should be "%d"" can be caused by code like:
xo_emit("{[:/%s}");
This code should be replaced with code like:
xo_emit("{[:/%d}");
Anchors only grok integer values, and if the value is not static, if must be in an 'int' argument, represented by the "%d" format. Anything else is an error.
The message "Max width only valid for strings" can be caused by code like:
xo_emit("{:tag/%2.4.6d}", 55);
This code should be replaced with code like:
xo_emit("{:tag/%2.6d}", 55);
libxo allows a true 'max width' in addition to the traditional printf-style 'max number of bytes to use for input'. But this is supported only for string values, since it makes no sense for non-strings. This error may occur from a typo, like "{:tag/%6..6d}" where only one period should be used.
This section provides task-oriented instructions for selected tasks. If you have a task that needs instructions, please open a request as an enhancement issue on github.
libxo is open source, under a new BSD license. Source code is available on github, as are recent releases. To get the most current release, please visit:
After downloading and untarring the source code, building involves the following steps:
sh bin/setup.sh
cd build
../configure
make
make test
sudo make install
libxo uses a distinct "build" directory to keep generated files separated from source files.
Use "../configure --help" to display available configuration options, which include the following:
--enable-warnings Turn on compiler warnings
--enable-debug Turn on debugging
--enable-text-only Turn on text-only rendering
--enable-printflike Enable use of GCC __printflike attribute
--disable-libxo-options Turn off support for LIBXO_OPTIONS
--with-gettext=PFX Specify location of gettext installation
--with-libslax-prefix=PFX Specify location of libslax config
Compiler warnings are a very good thing, but recent compiler version have added some very pedantic checks. While every attempt is made to keep libxo code warning-free, warnings are now optional. If you are doing development work on libxo, it is required that you use --enable-warnings to keep the code warning free, but most users need not use this option.
libxo provides the --enable-text-only option to reduce the footprint of the library for smaller installations. XML, JSON, and HTML rendering logic is removed.
The gettext library does not provide a simple means of learning its location, but libxo will look for it in /usr and /opt/local. If installed elsewhere, the installer will need to provide this information using the --with-gettext=/dir/path option.
libslax is not required by libxo; it contains the "oxtradoc" program used to format documentation.
To use libxo, you'll need to include the "xo.h" header file in your source code files:
#include <libxo/xo.h>
In your main() function, you'll need to call xo_parse_args to handling argument parsing (Parsing Command-line Arguments (xo_parse_args)). This function removes libxo-specific arguments the program's argv and returns either the number of remaining arguments or -1 to indicate an error.
int main (int argc, char **argv)
{
argc = xo_parse_args(argc, argv);
if (argc < 0)
return argc;
....
}
At the bottom of your main(), you'll need to call xo_finish() to complete output processing for the default handle (Handles). libxo provides the xo_finish_atexit function that is suitable for use with the atexit(3) function.
The second task is inspecting code for printf(3) calls and replacing them with xo_emit() calls. The format strings are similar in task, but libxo format strings wrap output fields in braces. The following two calls produce identical text output:
printf("There are %d %s events\n", count, etype);
xo_emit("There are {:count/%d} {:event} events\n", count, etype);
"count" and "event" are used as names for JSON and XML output. The "count" field uses the format "%d" and "event" uses the default "%s" format. Both are "value" roles, which is the default role.
Since text outside of output fields is passed verbatim, other roles are less important, but their proper use can help make output more useful. The "note" and "label" roles allow HTML output to recognize the relationship between text and the associated values, allowing appropriate "hover" and "onclick" behavior. Using the "units" role allows the presentation layer to perform conversions when needed. The "warning" and "error" roles allows use of color and font to draw attention to warnings. The "padding" role makes the use of vital whitespace more clear (The Padding Role ({P:})).
The "title" role indicates the headings of table and sections. This allows HTML output to use CSS to make this relationship more obvious.
Text output doesn't have any sort of hierarchy, but XML and JSON require this. Typically applications use indentation to represent these relationship:
printf("table %d\n", tnum);
for (i = 0; i < tmax; i++) {
printf(" %s %d\n", table[i].name, table[i].size);
}
xo_emit("{T:/table %d}\n", tnum);
xo_open_list("table");
for (i = 0; i < tmax; i++) {
xo_open_instance("table");
xo_emit("{P: }{k:name} {:size/%d}\n",
table[i].name, table[i].size);
xo_close_instance("table");
}
xo_close_list("table");
The open and close list functions are used before and after the list, and the open and close instance functions are used before and after each instance with in the list.
Typically these developer looks for a "for" loop as an indication of where to put these calls.
In addition, the open and close container functions allow for organization levels of hierarchy.
libxo provides variants of the standard error and warning functions, err(3) and warn(3). There are two variants, one for putting the errors on standard error, and the other writes the errors and warnings to the handle using the appropriate encoding style:
err(1, "cannot open output file: %s", file);
xo_err(1, "cannot open output file: %s", file);
xo_emit_err(1, "cannot open output file: {:filename}", file);
How do I use libxo to support internationalization?
libxo allows format and field strings to be used a keys into message catalogs to enable translation into a user's native language by invoking the standard gettext(3) functions.
gettext setup is a bit complicated: text strings are extracted from source files into "portable object template" (.pot) files using the "xgettext" command. For each language, this template file is used as the source for a message catalog in the "portable object" (.po) format, which are translated by hand and compiled into "machine object" (.mo) files using the "msgfmt" command. The .mo files are then typically installed in the /usr/share/locale or /opt/local/share/locale directories. At run time, the user's language settings are used to select a .mo file which is searched for matching messages. Text strings in the source code are used as keys to look up the native language strings in the .mo file.
Since the xo_emit format string is used as the key into the message catalog, libxo removes unimportant field formatting and modifiers from the format string before use so that minor formatting changes will not impact the expensive translation process. We don't want a developer change such as changing "/%06d" to "/%08d" to force hand inspection of all .po files. The simplified version can be generated for a single message using the "xopo -s <text>" command, or an entire .pot can be translated using the "xopo -f <input> -o <output>" command.
EXAMPLE:
% xopo -s "There are {:count/%u} {:event/%.6s} events\n"
There are {:count} {:event} events\n
Recommended workflow:
# Extract text messages
xgettext --default-domain=foo --no-wrap \
--add-comments --keyword=xo_emit --keyword=xo_emit_h \
--keyword=xo_emit_warn -C -E -n --foreign-user \
-o foo.pot.raw foo.c
# Simplify format strings for libxo
xopo -f foo.pot.raw -o foo.pot
# For a new language, just copy the file
cp foo.pot po/LC/my_lang/foo.po
# For an existing language:
msgmerge --no-wrap po/LC/my_lang/foo.po \
foo.pot -o po/LC/my_lang/foo.po.new
# Now the hard part: translate foo.po using tools
# like poedit or emacs' po-mode
# Compile the finished file; Use of msgfmt's "-v" option is
# strongly encouraged, so that "fuzzy" entries are reported.
msgfmt -v -o po/my_lang/LC_MESSAGES/foo.mo po/my_lang/foo.po
# Install the .mo file
sudo cp po/my_lang/LC_MESSAGES/foo.mo \
/opt/local/share/locale/my_lang/LC_MESSAGE/
Once these steps are complete, you can use the "gettext" command to test the message catalog:
Note the field names do not change and they should not be translated. The contents of the note ("byte,bytes") should also not be translated, since the "g" modifier will need the untranslated value as the key for the message catalog.
The field "{g:server}" requests the rendered value of the field be translated using gettext(3). In this example, "web" would be used.
The field "{Ngp:byte,bytes}" shows an example of plural form using the "p" modifier with the "g" modifier. The base singular and plural forms appear inside the field, separated by a comma. At run time, libxo uses the previous field's numeric value to decide which form to use by calling ngettext(3).
If a domain name is needed, it can be supplied as the content of the {G:} role. Domain names remain in use throughout the format string until cleared with another domain name.
printf(dgettext("dns", "Host %s not found: %d(%s)\n"),
name, errno, dgettext("strerror", strerror(errno)));
xo_emit("{G:dns}Host {:hostname} not found: "
"%d({G:strerror}{g:%m})\n", name, errno);
% ./testxo --libxo text
Item 'gum':
Total sold: 1412.0
In stock: 54
On order: 10
SKU: GRO-000-415
Item 'rope':
Total sold: 85.0
In stock: 4
On order: 2
SKU: HRD-000-212
Item 'ladder':
Total sold: 0
In stock: 2
On order: 1
SKU: HRD-000-517
Item 'bolt':
Total sold: 4123.0
In stock: 144
On order: 42
SKU: HRD-000-632
Item 'water':
Total sold: 17.0
In stock: 14
On order: 2
SKU: GRO-000-2331
Item 'fish':
Total sold: 1321.0
In stock: 45
On order: 1
SKU: GRO-000-533