[TOC]
The following primitive types are found within input files:
'single quotes'
or "double quotes"
. By convention, single
quotes are used.1
and false with 0
.Lists, which are represented as a sequence of items separated by
commas (,
) within square brackets ([
and ]
). A list may
contain any other primitive types, including other lists.
Generally, each item of a list must be of the same type as all other
items in the list, but in some cases (such as within conditions
sections), the list structure is more tightly specified. A trailing
comma is permitted.
This example list contains three string values.
[ 'Generate', 'Your', 'Projects', ]
Dictionaries, which map keys to values. All keys are strings.
Values may be of any other primitive type, including other
dictionaries. A dictionary is enclosed within curly braces ({
and
}
). Keys precede values, separated by a colon (:
). Successive
dictionary entries are separated by commas (,
). A trailing comma
is permitted. It is an error for keys to be duplicated within a
single dictionary as written in an input file, although keys may
replace other keys during merging.
This example dictionary maps each of three keys to different values.
{
'inputs': ['version.c.in'],
'outputs': ['version.c'],
'process_outputs_as_sources': 1,
}
A GYP input file is organized as structured data. At the root scope of
each .gyp
or .gypi
(include) file is a dictionary. The keys and
values of this dictionary, along with any descendants contained within
the values, provide the data contained within the file. This data is
given meaning by interpreting specific key names and their associated
values in specific ways (see Settings Keys).
Within an input file, a comment is introduced by a pound sign (#
) not
within a string. Any text following the pound sign, up until the end of
the line, is treated as a comment.
{
'school_supplies': [
'Marble composition book',
'Sharp #2 pencil',
'Safety scissors', # You still shouldn't run with these
],
}
In this example, the # in 'Sharp #2 pencil'
is not taken as
introducing a comment because it occurs within a string, but the text
after 'Safety scissors'
is treated as a comment having no impact on
the data within the file.
Many operations on GYP input files occurs by merging dictionary and list items together. During merge operations, it is important to recognize the distinction between source and destination values. Items from the source value are merged into the destination, which leaves the source unchanged and the destination modified by the source. A dictionary may only be merged into another dictionary, and a list may only be merged into another list.
=
), the policy is for the
source list to completely replace the destination list if it
exists. Mnemonic: =
for assignment.?
), the policy is for the
source list to be set as the destination list only if the key is
not already present in the destination. Mnemonic: ?
for
conditional assignment.+
), the policy is for the
source list contents to be prepended to the destination list.
Mnemonic: +
for addition or concatenation.Source dictionary:
{
'include_dirs+': [
'shared_stuff/public',
],
'link_settings': {
'libraries': [
'-lshared_stuff',
],
},
'test': 1,
}
Destination dictionary:
{
'target_name': 'hello',
'sources': [
'kitty.cc',
],
'include_dirs': [
'headers',
],
'link_settings': {
'libraries': [
'-lm',
],
'library_dirs': [
'/usr/lib',
],
},
'test': 0,
}
Merged dictionary:
{
'target_name': 'hello',
'sources': [
'kitty.cc',
],
'include_dirs': [
'shared_stuff/public', # Merged, list item prepended due to include_dirs+
'headers',
],
'link_settings': {
'libraries': [
'-lm',
'-lshared_stuff', # Merged, list item appended
],
'library_dirs': [
'/usr/lib',
],
},
'test': 1, # Merged, int value replaced
}
In a .gyp
or .gypi
file, many string values are treated as pathnames
relative to the file in which they are defined.
String values associated with the following keys, or contained within lists associated with the following keys, are treated as pathnames:
Additionally, string values associated with keys ending in the following suffixes, or contained within lists associated with keys ending in the following suffixes, are treated as pathnames:
_dir
_dirs
_file
_files
_path
_paths
However, any string value beginning with any of these characters is excluded from pathname relativization:
/
for identifying absolute paths.$
for introducing build system variable expansions.-
to support specifying such items as -llib
, meaning “library
lib
in the library search path.”<
, >
, and !
for GYP expansions.When merging such relative pathnames, they are adjusted so that they can remain valid relative pathnames, despite being relative to a new home.
Source dictionary from ../build/common.gypi
:
{
'include_dirs': ['include'], # Treated as relative to ../build
'libraries': ['-lz'], # Not treated as a pathname, begins with a dash
'defines': ['NDEBUG'], # defines does not contain pathnames
}
Target dictionary, from base.gyp
:
{
'sources': ['string_util.cc'],
}
Merged dictionary:
{
'sources': ['string_util.cc'],
'include_dirs': ['../build/include'],
'libraries': ['-lz'],
'defines': ['NDEBUG'],
}
Because of pathname relativization, after the merge is complete, all of
the pathnames in the merged dictionary are valid relative to the
directory containing base.gyp
.
Some list items are treated as singletons, and the list merge process
will enforce special rules when merging them. At present, any string
item in a list that does not begin with a dash (-
) is treated as a
singleton, although this is subject to change. When appending or
prepending a singleton to a list, if the item is already in the list,
only the earlier instance is retained in the merged list.
Source dictionary:
{
'defines': [
'EXPERIMENT=1',
'NDEBUG',
],
}
Destination dictionary:
{
'defines': [
'NDEBUG',
'USE_THREADS',
],
}
Merged dictionary:
{
'defines': [
'NDEBUG',
'USE_THREADS',
'EXPERIMENT=1', # Note that NDEBUG is not appended after this.
],
}
If the -I
(--include
) argument was used to invoke GYP, any files
specified will be implicitly merged into the root dictionary of all
.gyp
files.
An includes section may be placed anywhere within a
.gyp
or .gypi
(include) file. includes
sections contain lists of
other files to include. They are processed sequentially and merged into
the enclosing dictionary at the point that the includes
section was
found. includes
sections at the root of a .gyp
file dictionary are
merged after any -I
includes from the command line.
includes sections are processed immediately after a file is loaded, even before variable and conditional processing, so it is not possible to include a file based on a variable reference. While it would be useful to be able to include files based on variable expansions, it is most likely more useful to allow included files access to variables set by the files that included them.
An includes section may, however, be placed within a conditional section. The included file itself will be loaded unconditionally, but its dictionary will be discarded if the associated condition is not true.
There are three main types of variables within GYP.
CAPITAL_LETTERS
. Predefined variables are set automatically by
GYP. They may be overridden, but it is not advisable to do so. See
Predefined Variables for a list of
variables that GYP provides.variables
can be provided, containing a mapping between variable
names (keys) and their contents (values), which may be strings,
integers, or lists of strings. By convention, user-defined
variables are named with lowercase_letters
._
) prefixed. For example, if
your dictionary contains type: 'static_library'
, an automatic
variable named _type
will be provided, and its value will be a
string, 'static_library'
.Variables are inherited from enclosing scopes.
Within a variables
section, keys named with percent sign (%
)
suffixes mean that the variable should be set only if it is undefined at
the time it is processed. This can be used to provide defaults for
variables that would otherwise be undefined, so that they may reliably
be used in variable expansion or conditional
processing.
Each GYP generator module provides defaults for the following variables:
OS
: The name of the operating system that the generator produces
output for. Common values for values for OS
are:
'linux'
'mac'
'win'
But other values may be encountered and this list should not be
considered exhaustive. The gypd
(debug) generator module does not
provide a predefined value for OS
. When invoking GYP with the
gypd
module, if a value for OS
is needed, it must be provided on
the command line, such as gyp -f gypd -DOS=mac
.
GYP generators also provide defaults for these variables. They may
be expressed in terms of variables used by the build system that
they generate for, often in $(VARIABLE)
format. For example, the
GYP PRODUCT_DIR
variable maps to the Xcode BUILT_PRODUCTS_DIR
variable, so PRODUCT_DIR
is defined by the Xcode generator as
$(BUILT_PRODUCTS_DIR)
.
EXECUTABLE_PREFIX
: A prefix, if any, applied to executable names.
Usually this will be an empty string.EXECUTABLE_SUFFIX
: A suffix, if any, applied to executable names.
On Windows, this will be .exe
, elsewhere, it will usually be an
empty string.INTERMEDIATE_DIR
: A directory that can be used to place
intermediate build results in. INTERMEDIATE_DIR
is only
guaranteed to be accessible within a single target (See targets).
This variable is most useful within the context of rules and actions
(See rules, See actions). Compare with SHARED_INTERMEDIATE_DIR
.PRODUCT_DIR
: The directory in which the primary output of each
target, such as executables and libraries, is placed.RULE_INPUT_ROOT
: The base name for the input file (e.g. “foo
”).
See Rules.RULE_INPUT_EXT
: The file extension for the input file (e.g.
“.cc
”). See Rules.RULE_INPUT_NAME
: Full name of the input file (e.g. “foo.cc
”).
See Rules.RULE_INPUT_PATH
: Full path to the input file (e.g.
“/bar/foo.cc
”). See Rules.SHARED_INTERMEDIATE_DIR
: A directory that can be used to place
intermediate build results in, and have them be accessible to other
targets. Unlike INTERMEDIATE_DIR
, each target in a project,
possibly spanning multiple .gyp
files, shares the same
SHARED_INTERMEDIATE_DIR
.The following additional predefined variables may be available under certain circumstances:
DEPTH
. When GYP is invoked with a --depth
argument, when
processing any .gyp
file, DEPTH
will be a relative path from the
.gyp
file to the directory specified by the --depth
argument.A user-defined variable may be defined in terms of other variables, but not other variables that have definitions provided in the same scope.
GYP provides two forms of variable expansions, “early” or “pre” expansions, and “late,” “post,” or “target” expansions. They have similar syntax, differing only in the character used to introduce them.
<
) character.
Mnemonic: the arrow points to the left, earlier on a timeline.>
) character.
Mnemonic: the arrow points to the right, later on a timeline.The difference the two phases of expansion is described in Early and Late Phases.
These characters were chosen based upon the requirement that they not
conflict with the variable format used natively by build systems. While
the dollar sign ($
) is the most natural fit for variable expansions,
its use was ruled out because most build systems already use that
character for their own variable expansions. Using different characters
means that no escaping mechanism was needed to differentiate between GYP
variables and build system variables, and writing build system variables
into GYP files is not cumbersome.
Variables may contain lists or strings, and variable expansions may occur in list or string context. There are variant forms of variable expansions that may be used to determine how each type of variable is to be expanded in each context.
<(VAR)
or >(VAR)
:
VAR
is a string, the variable reference within the string is
replaced by variable’s string value.VAR
is a list, the variable reference within the string is
replaced by a string containing the concatenation of all of the
variable’s list items. Generally, the items are joined with
spaces between each, but the specific behavior is
generator-specific. The precise encoding used by any generator
should be one that would allow each list item to be treated as a
separate argument when used as program arguments on the system
that the generator produces output for.<@(VAR)
or >@(VAR)
:
'<@(VAR)'
or '>@(VAR)'
exactly.VAR
is a list, each of its elements are inserted into the
list in which expansion is taking place, replacing the list item
containing the variable reference.VAR
is a string, the string is converted to a list which is
inserted into the list in which expansion is taking place as
above. The conversion into a list is generator-specific, but
generally, spaces in the string are taken as separators between
list items. The specific method of converting the string to a
list should be the inverse of the encoding method used to expand
list variables in string context, above.GYP treats references to undefined variables as errors.
Command expansions function similarly to variable expansions, but
instead of resolving variable references, they cause GYP to execute a
command at generation time and use the command’s output as the
replacement. Command expansions are introduced by a less than and
exclamation mark (<!
).
In a command expansion, the entire string contained within the
parentheses is passed to the system’s shell. The command’s output is
assigned to a string value that may subsequently be expanded in list
context in the same way as variable expansions if an @
character is
used.
In addition, command expansions (unlike other variable expansions) may include nested variable expansions. So something like this is allowed:
'variables' : {
'foo': '<!(echo Build Date <!(date))',
},
expands to:
'variables' : {
'foo': 'Build Date 02:10:38 PM Fri Jul 24, 2009 -0700 PDT',
},
You may also put commands into arrays in order to quote arguments (but note that you need to use a different string quoting character):
'variables' : {
'files': '<!(["ls", "-1", "Filename With Spaces"])',
},
GYP treats command failures (as indicated by a nonzero exit status) during command expansion as errors.
You can use the special pymod_do_main
symbol to get GYP to import the specified Python module, and call it’s DoMain
method passing any parameters given.
{
'variables' : {
'res': '<!pymod_do_main(modulename par ame ters)',
},
}
{
'sources': [
'<!(echo filename with space.cc)',
],
'libraries': [
'<!@(pkg-config --libs-only-l apr-1)',
],
}
might expand to:
{
'sources': [
'filename with space.cc', # no @, expands into a single string
],
'libraries': [ # @ was used, so there's a separate list item for each lib
'-lapr-1',
'-lpthread',
],
}
Conditionals use the same set of variables used for variable expansion. As with variable expansion, there are two phases of conditional evaluation:
The syntax for each type is identical, they differ only in the key name used to identify them and the timing of their evaluation. A more complete description of syntax and use is provided in conditions.
The difference the two phases of evaluation is described in Early and Late Phases.
GYP performs two phases of variable expansion and conditional evaluation:
<
form of variable expansions.>
form
of variable expansions,
and on the !
form of command expansions.These two phases are provided because there are some circumstances in which each is desirable.
The “early” phase is appropriate for most expansions and evaluations.
“Early” expansions and evaluations may be performed anywhere within any
.gyp
or .gypi
file.
The “late” phase is appropriate when expansion or evaluation must be
deferred until a specific section has been merged into target context.
“Late” expansions and evaluations only occur within targets
sections
and their descendants. The typical use case for a late-phase expansion
is to provide, in some globally-included .gypi
file, distinct
behaviors depending on the specifics of a target.
Given this input:
{
'target_defaults': {
'target_conditions': [
['_type=="shared_library"', {'cflags': ['-fPIC']}],
],
},
'targets': [
{
'target_name': 'sharing_is_caring',
'type': 'shared_library',
},
{
'target_name': 'static_in_the_attic',
'type': 'static_library',
},
]
}
The conditional needs to be evaluated only in target context; it is
nonsense outside of target context because no _type
variable is
defined. target_conditions allows evaluation
to be deferred until after the targets sections are
merged into their copies of target_defaults.
The resulting targets, after “late” phase processing:
{
'targets': [
{
'target_name': 'sharing_is_caring',
'type': 'shared_library',
'cflags': ['-fPIC'],
},
{
'target_name': 'static_in_the_attic',
'type': 'static_library',
},
]
}
During any expansion and evaluation phase, both expansion and evaluation are performed simultaneously. The process for handling variable expansions and conditional evaluation within a dictionary is:
variables
dictionary.One quirk of this ordering is that you cannot expect a variables section within a dictionary’s conditional to be effective in the dictionary itself, but the added variables will be effective in any child dictionaries or lists. It is thought to be far more worthwhile to provide resolved automatic variables to conditional sections, though. As a workaround, to conditionalize variable values, place a conditions or target_conditions section within the variables section.
In GYP, “dependents” are targets that rely on other targets, called “dependencies.” Dependents declare their reliance with a special section within their target dictionary, dependencies.
It is useful for targets to “advertise” settings to their dependents. For example, a target might require that all of its dependents add certain directories to their include paths, link against special libraries, or define certain preprocessor macros. GYP allows these cases to be handled gracefully with “dependent settings” sections. There are three types of such sections:
Furthermore, in some cases, a target needs to pass its dependencies’ settings on to its own dependents. This might happen when a target’s own public header files include header files provided by its dependency. export_dependent_settings allows a target to declare dependencies for which direct_dependent_settings should be passed through to its own dependents.
Dependent settings processing merges a copy of the relevant dependent settings dictionary from a dependency into its relevant dependent targets.
In most instances, direct_dependent_settings will be used. There are very few cases where all_dependent_settings is actually correct; in most of the cases where it is tempting to use, it would be preferable to declare export_dependent_settings. Most libraries and library_dirs sections should be placed within link_settings sections.
Given:
{
'targets': [
{
'target_name': 'cruncher',
'type': 'static_library',
'sources': ['cruncher.cc'],
'direct_dependent_settings': {
'include_dirs': ['.'], # dependents need to find cruncher.h.
},
'link_settings': {
'libraries': ['-lm'], # cruncher.cc does math.
},
},
{
'target_name': 'cruncher_test',
'type': 'executable',
'dependencies': ['cruncher'],
'sources': ['cruncher_test.cc'],
},
],
}
After dependent settings processing, the dictionary for cruncher_test
will be:
{
'target_name': 'cruncher_test',
'type': 'executable',
'dependencies': ['cruncher'], # implies linking against cruncher
'sources': ['cruncher_test.cc'],
'include_dirs': ['.']
'libraries': ['-lm'],
},
If cruncher
was declared as a shared_library
instead of a
static_library
, the cruncher_test
target would not contain -lm
,
but instead, cruncher
itself would link against -lm
.
The precise meaning of a dependency relationship varies with the types of the targets at either end of the relationship. In GYP, a dependency relationship can indicate two things about how targets relate to each other:
The analysis of the first item is complicated by the differences between static and shared libraries.
.o
or
.obj
) that are used as inputs to a linker (ld
or link.exe
).
Static libraries don’t link against other libraries, they’re
collected together and used when eventually linking a shared library
or executable.Accordingly, GYP performs an operation referred to as “static library dependency adjustment,” in which it makes each linker output target (shared libraries and executables) link against the static libraries it depends on, either directly or indirectly. Because the linkable targets link against these static libraries, they are also made direct dependents of the static libraries.
As part of this process, GYP is also able to remove the direct dependency relationships between two static library targets, as a dependent static library does not actually need to link against a dependency static library. This removal facilitates speedier builds under some build systems, as they are now free to build the two targets in parallel. The removal of this dependency is incorrect in some cases, such as when the dependency target contains rules or actions that generate header files required by the dependent target. In such cases, the dependency target, the one providing the side-effect files, must declare itself as a hard_dependency. This setting instructs GYP to not remove the dependency link between two static library targets in its generated output.
When GYP runs, it loads all .gyp
files needed to resolve dependencies
found in dependencies sections. These files are not
merged into the files that reference them, but they may contain special
sections that are merged into dependent target dictionaries.
Explain this.
GYP allows list items to be filtered by “exclusions” and “patterns.” Any list containing string values in a dictionary may have this filtering applied. For the purposes of this section, a list modified by exclusions or patterns is referred to as a “base list”, in contrast to the “exclusion list” and “pattern list” that operates on it.
key
, the key!
list
provides exclusions.key
, the key/
list
provides regular expression pattern-based filtering.Both key!
and key/
may be present. The key!
exclusion list will
be processed first, followed by the key/
pattern list.
Exclusion lists are most powerful when used in conjunction with conditionals.
An exclusion list provides a way to remove items from the related list based on exact matching. Any item found in an exclusion list will be removed from the corresponding base list.
This example excludes files from the sources
based on the setting of
the OS
variable.
{
'sources:' [
'mac_util.mm',
'win_util.cc',
],
'conditions': [
['OS=="mac"', {'sources!': ['win_util.cc']}],
['OS=="win"', {'sources!': ['mac_util.cc']}],
],
}
Pattern lists are similar to, but more powerful than, exclusion
lists. Each item in a pattern list is itself
a two-element list. The first item is a string, either 'include'
or
'exclude'
, specifying the action to take. The second item is a string
specifying a regular expression. Any item in the base list matching the
regular expression pattern will either be included or excluded, based on
the action specified.
Items in a pattern list are processed in sequence, and an excluded item that is later included will not be removed from the list (unless it is subsequently excluded again.)
Pattern lists are processed after exclusion lists, so it is possible for a pattern list to re-include items previously excluded by an exclusion list.
Nothing is actually removed from a base list until all items in an exclusion list and pattern list have been evaluated. This allows items to retain their correct position relative to one another even after being excluded and subsequently included.
In this example, a uniform naming scheme is adopted for platform-specific files.
{
'sources': [
'io_posix.cc',
'io_win.cc',
'launcher_mac.cc',
'main.cc',
'platform_util_linux.cc',
'platform_util_mac.mm',
],
'sources/': [
['exclude', '_win\\.cc$'],
],
'conditions': [
['OS!="linux"', {'sources/': [['exclude', '_linux\\.cc$']]}],
['OS!="mac"', {'sources/': [['exclude', '_mac\\.cc|mm?$']]}],
['OS=="win"', {'sources/': [
['include', '_win\\.cc$'],
['exclude', '_posix\\.cc$'],
]}],
],
}
After the pattern list is applied, sources
will have the following
values, depending on the setting of OS
:
OS
is linux
: ['io_posix.cc', 'main.cc',
'platform_util_linux.cc']
OS
is mac
: ['io_posix.cc', 'launcher_mac.cc', 'main.cc',
'platform_util_mac.mm']
OS
is win
: ['io_win.cc', 'main.cc',
'platform_util_win.cc']
Note that when OS
is win
, the include
for _win.cc
files is
processed after the exclude
matching the same pattern, because the
sources/
list participates in merging during
conditional evaluation just like any other list
would. This guarantees that the _win.cc
files, previously
unconditionally excluded, will be re-included when OS
is win
.
In some cases, a GYP generator needs to access to items that were
excluded by an exclusion list or pattern
list. When GYP excludes items during processing
of either of these list types, it places the results in an _excluded
list. In the example above, when OS
is mac
, sources_excluded
would be set to ['io_win.cc', 'platform_util_linux.cc']
. Some GYP
generators use this feature to display excluded files in the project
files they generate for the convenience of users, who may wish to refer
to other implementations.
GYP uses a defined and predictable order to execute the various steps performed between loading files and generating output.
.gyp
files. Merge any command-line
includes into each .gyp
file’s root
dictionary. As includes are found,
load them as well and merge them into the scope in
which the includes section was found..gyp
file’s root target_defaults
dictionary..gyp
files not yet loaded.List of condition
items
A conditions
section introduces a subdictionary that is only merged
into the enclosing scope based on the evaluation of a conditional
expression. Each condition
within a conditions
list is itself a
list of at least two items:
var=="value"
and var!="value"
to test
equality and inequality. For example, 'OS=="linux"'
is true
when the OS
variable is set to "linux"
.
* For integer values, var==value
, var!=value
, var<value
,
var<=value
, var>=value
, and var>value
, to test equality and
several common forms of inequality. For example,
'chromium_code==0'
is true when the chromium_code
variable is
set to 0
.
* It is an error for a conditional expression to reference any
undefined variable.These two items can be followed by any number of similar two items that will be evaluated if the previous conditional expression does not evaluate to true.
An additional optional dictionary can be appended to this sequence of two items. This optional dictionary will be merged into the enclosing scope if none of the conditional expressions evaluate to true.
Within a conditions
section, each item is processed sequentially, so
it is possible to predict the order in which operations will occur.
There is no restriction on nesting conditions
sections.
conditions
sections are very similar to target_conditions
sections.
See target_conditions.
{
'sources': [
'common.cc',
],
'conditions': [
['OS=="mac"', {'sources': ['mac_util.mm']}],
['OS=="win"', {'sources': ['win_main.cc']}, {'sources': ['posix_main.cc']}],
['OS=="mac"', {'sources': ['mac_impl.mm']},
'OS=="win"', {'sources': ['win_impl.cc']},
{'sources': ['default_impl.cc']}
],
],
}
Given this input, the sources
list will take on different values based
on the OS
variable.
OS
is "mac"
, sources
will contain ['common.cc',
'mac_util.mm', 'posix_main.cc', 'mac_impl.mm']
.OS
is "win"
, sources
will contain ['common.cc',
'win_main.cc', 'win_impl.cc']
.OS
is any other value such as "linux"
, sources
will contain
['common.cc', 'posix_main.cc', 'default_impl.cc']
.