Speeding up Phan Analysis - TysonAndre/phan GitHub Wiki
The following suggestions may help speed up Phan analysis on your project:
-
Make sure nothing is slowing down the PHP interpreter. (e.g. xdebug)
-
Avoid using known slow Phan configuration operations for CI for large projects (and projects with large vendor directories). You can periodically run dead code checks, etc. manually.
-
Reduce the number of files phan parses. Phan ignores any PHP autoloading configuration that is set up. It unconditionally parses all files and directories listed in the parsed file and directory list.
-
Run Phan only on the files you changed (Daemon mode or --include-analysis-file-list). This may be useful during local development, but not for Continuous Integration.
-
Disable xdebug before running Phan (Phan runs around 5 times slower with xdebug enabled)
The output of
php -mwill mention xdebug if xdebug is still enabled.Xdebug may be disabled in in various ways, e.g. by setting the environment variable
PHPRCto the absolute path to a folder containing a copy of php.ini without xdebug.(XDebug is only suggested if you're investigating a crash, or a test failure in Phan's test suite, or an uncaught error in Phan, and shouldn't be used in normal usage.)
-
Run
php --version. Normally, you should see (NTS), not (ZTS DEBUG) or (NTS DEBUG), but if you build PHP yourself (e.g. for PECL module development), the version you're currently using may have been built with--enable-debug(DEBUG). php built with--enable-debug(DEBUG) is around 2 times slower.
These configuration options should be changed to the below values in your project's .phan/config.php. (For the settings with a value of false, omitting them would do the same thing)
// Backwards Compatibility Checking (Disable this if the application no longer supports php 5, or use a different tool. Phan's checks are currently slow)
// Set it to false or omit it.
'backward_compatibility_checks' => false,
// A regular expression to match files to be excluded
// from parsing and analysis and will not be read at all.
//
// This is useful for excluding groups of test or example
// directories/files, unanalyzable files, or files that
// can't be removed for whatever reason.
// (e.g. '@Test\.php$@', or '@vendor/.*/(tests|Tests)/@')
//
'exclude_file_regex' => '@^vendor/.*/(tests|Tests)/@',
// This hasn't been optimized yet, and is still experimental, but optimization is planned.
// Set it to false or omit it.
'simplify_ast' => false,
// This is somewhat slow and doesn't work with multiple processes.
// Set it to false or omit it.
'dead_code_detection' => false,Phan currently ignores any autoloading configuration that is set up for your project. It assumes that the files will be autoloaded correctly, and will unconditionally parses all files and directories listed in the parsed file and directory list (except for the excluded ones).
For example, if your project's .phan/config have config options similar to the below settings:
// An example of a config which parses too many files:
'directory_list' => [
'src',
'tests', // This includes more files than needed, and tests may have high false positives rates
'vendor', // This includes more files than needed. Only include direct dependencies
'path/to/NonComposerThirdPartyProjects',
],
"exclude_analysis_directory_list" => [
'vendor/'
'tests/'
'path/to/NonComposerThirdPartyProjects',
],Then it would be faster to change it to the below (may require some tweaking):
// Be as specific as possible with 'directory_list' and 'file_list' to avoid parsing files unnecessarily.
'directory_list' => [
'src',
'tests/ClassesUsedBySrc/', // If the tests pass, you may not need to run Phan on your unit tests. However, some files may still need to be included, e.g. for constant definitions, defining classes used by `instanceof` checks, etc.
// You can parse fewer files if you list only direct dependencies (exclude as many devDependencies as possible). (And maybe dependencies of those dependencies, to fix any Phan issues that show up
// Additionally, it's faster to avoid parsing tests, examples, and
'vendorName/directDependencyProject1/src',
'vendorName/directDependencyProject2/library',
'vendorName/directDependencyProject3/src/ProjectName',
'path/to/NonComposerThirdPartyProjects/project1/src',
'path/to/NonComposerThirdPartyProjects/project2/src',
],
// No change here
"exclude_analysis_directory_list" => [
'vendor/'
'tests/'
'path/to/NonComposerThirdPartyProjects',
],
// A regular expression to match files to be excluded
// from parsing and analysis and will not be read at all.
//
// This is useful for excluding groups of test or example
// directories/files, unanalyzable files, or files that
// can't be removed for whatever reason.
// (e.g. '@Test\.php$@', or '@vendor/.*/(tests|Tests)/@')
'exclude_file_regex' => '@^(vendor|path/to/NonComposerThirdPartyProjects)/.*/(tests|Tests|doc|examples)/@',The analysis phase generally takes much longer than the parse phase.
Note that this will significantly increase the total amount of memory Phan will use (Watch the memory usage with a tool such as htop (unix/linux)).
If you're planning to start using multiple processes and haven't done so before, it's a good idea to read Different Issue Sets On Different Numbers of CPUs and what to do about different issue sets
This will be less accurate, but much faster (Won't be able to infer types of unspecified parameters from calls made from outside of the function).
There are two ways to do this.
- By Using Phan Daemon Mode (Experimental), which lets you reuse the results of the parse phase
- By invoking Phan with
--include-analysis-file-list
--include-analysis-file-list <file_list>
A comma-separated list of files that will be included in
static analysis. All others won't be analyzed.
This is primarily intended for performing standalone
incremental analysis.
Phan 0.9.2 (For php 7.1) contains many small performance optimizations, and more optimizations are planned for the next release.
Those optimizations were backported to the 0.8.4 release (for php 7.0).