Parse Avoidance

8 minute read

CloudBees Build Acceleration includes a parse avoidance feature that can almost eliminate makefile parse time. By caching and reusing parse result files, CloudBees Build Acceleration can speed up both full builds and incremental builds.

Parse avoidance works when emulating GNU Make with clusters or local agents.

Parse avoidance uses the concept of cache "slots." When parse avoidance is enabled, eMake maintains a slot for each combination of non-ignored command line options, non-ignored environment variable assignments, and current working directory. (Ignored arguments and environment variables are listed below.) A slot can be empty or hold a previously cached result. If the appropriate slot holds an up-to-date result, parsing is avoided.

A cached result becomes obsolete if eMake detects file system changes that might have caused a different result (with the same command line options, environment variable assignments, and current working directory). Such file system changes include any file read by the parse job, which means all makefiles, all programs invoked by $(shell) during the parse (as opposed to during rule execution), the files they read, and so on.

Console Output

When a cached parse result is reused, eMake replays the file system modifications made during the parse job and any standard output or standard error that the original parse job produced. For example, if the makefile includes

VARIABLE1 := $(shell echo Text > myfile)
VARIABLE2 := $(warning Warning: updated myfile)

then when using a cached parse result, eMake creates "myfile" and prints "Warning: updated myfile".

Enabling Parse Avoidance

You must first run a "learning" build with the parse avoidance feature enabled. To enable parse avoidance, set the following: --emake-parse-avoidance=1

For the learning build, (because the parse cache is empty), the argument only saves a new result to the cache. For subsequent builds, the argument enables the reuse of cached parse results and saves a new result to the cache as appropriate. If you do not specify --emake-parse-avoidance=1, then the parse avoidance cache is not accessed at all.

You should also disable generated dependencies (either by modifying your makefiles or by using --emake-suppress-include ; see below)

The following table describes parse avoidance-related options.

eMake Option Description

--emake-assetdir=< path >

Use the specified directory for cached parse results. The default directory is named .emake. (This option also determines the location of the saved dependency information for the dependency optimization feature and the cache location for JobCache .)

--emake-parse-avoidance= < 0/1 >

Avoid parsing makefiles when prior parse results remain up-to-date and cache new parse results when appropriate.

--emake-parse-avoidance-ignore-env=< var >

Ignore the named environment variable when searching for applicable cached parse results. To ignore more than one variable, use this option multiple times.

--emake-parse-avoidance-ignore-path=< path >

Ignore this file or directory when checking whether cached parse results are up-to-date. Append % for prefix matching (the % must be the last character). To ignore more than one path or prefix, use this option multiple times.

Incorrect placement of % will result in an error.

Correct: --emake-parse-avoidance-ignore-path=.foo%

Incorrect: --emake-parse-avoidance-ignore-path=.%foo

--emake-suppress-include= < pattern >

Skip matching makefile includes (such as generated dependencies). Generally, you should not suppress makefile includes unless they are generated dependency files, and you have enabled automatic dependencies as an alternative way of handling dependencies.

If the pattern does not have a directory separator, then the pattern is compared to the include’s file name component only. If the pattern has a directory separator, then the pattern is taken relative to the same working directory that applies to the include directive and compared to the included file’s entire path.

If a file is read during a parse but changes before eMake attempts to reuse that parse’s results, the cached parse result is normally considered to be obsolete. You can, however, temporarily override this decision with --emake-parse-avoidance-ignore-path.

eMake permanently ignores the effect on a given parse result of certain special files that might have existed when it was created, in all cases using the names they would have had at that time:

  • Anything in ".emake" or whatever alternative directory you specified using --emake-assetdir

  • Client-side eMake debug log (as specified by --emake-logfile )

  • Annotation file (as specified by --emake-annofile )

  • History file (as specified by --emake-historyfile )

  • Your makefiles should not contain any $(shell) expansions that modify these four special files and directories in ways that affect the build, because parse avoidance will ignore changes to them.

  • Enabling remote parse debugging (the --emake-rdebug option) disables parse avoidance.

You can use a pragma to instruct parse avoidance to detect the dependence of a parse result upon path wildcard match results. This pragma makes the parse cache sensitive to the existence (or nonexistence) of specific files in directories that it reads as well as in new subdirectories thereof. (eMake does not detect matching files in just any new subdirectory, but only those that are subdirectories of directories read in the original job and their subdirectories, recursively.)

You can specify more than one glob pattern, as in the following pragma: #pragma jobcache parse -readdir *.h -readdir *.c

For Android-specific information about parse avoidance and other Android best practices, see Optimizing Android Build Performance.

Parse Avoidance Example

noautodep.mk is used for this example. The contents are:

#pragma noautodep */.git/* $(local-intermediates-dir)/libbcc-stamp.c :
#pragma noautodep */out/target/product/generic/system/bin/cat $(linked_module) :

The following parse avoidance example provides command line arguments for Android 4.1.1.

You might need additional options such as ` --emake-cm`.
--emake-parse-avoidance=1 --emake-autodepend=1 --emake-suppress-include=*.d --emake-suppress-include=*.P --emake-debug=P -f Makefile -f noautodep.mk

Deleting the Cache

To delete the cache, delete <assetdir>/cache.*. (The default asset directory is .emake.)

For example:

rm -r .emake/cache.*

Moving Your Workspace

If you want to move your workspace, make sure that the new eMake roots correspond to the old eMake roots. Also, because the asset directory defaults to .emake in the current working directory, you must either copy that directory to the new workspace or use --emake-assetdir= to specify an asset directory that you want the two workspaces to share. If you already use --emake-assetdir= to point to an asset directory within your old workspace and also want to move the asset directory, you must update its value to point to the new asset directory location.

eMake looks for eMake roots in the values of Make variables. When eMake replays a cached parse result in a new workspace, it replaces the old eMake roots in the values of those variables with the new eMake roots. This policy works well as long as no confusion exists between eMake roots and other text in those variable values. For example, if the value of a Make variable is -I/we will look into this, your old eMake root is /we, and your new eMake root is /wg, then the new value will be -I/wg will look into this. For best results, choose distinctive directory names for your workspaces.

Parse Avoidance Limitations

  • Windows is currently not supported.

  • The parse avoidance feature does not detect the need to reparse in these cases:

  • Changes to input files and programs that are used during the parse (such as makefile includes) but are not virtualized (because they are not located under --emake-root=…​ ).

  • Changes to non file system, non-registry aspects of the environment, for example, the current time of day

  • If a parse job checks the existence or timestamp of a file without reading it, parse avoidance might not invalidate its cached result when the file is created, destroyed, or modified.

  • Using --emake-autodepend=1 and --emake-suppress-include=<pattern> in conjunction with parse avoidance helps to avoid this limitation. If a generated dependency file did not exist when a parse result was saved into the cache, eMake might reuse that cached parse result even after that dependency file was created. Of course, you also benefit from eDepend’s performance and reliability gains.

  • This limitation can be mitigated by the use of #pragma jobcache parse -readdir …​, which reruns the parse job after files are created or deleted that match the pattern and are located in a directory that was subject to a wildcard match or $(shell find …​) when the cache was populated.

Troubleshooting

Enabling Parse Avoidance Debug Logging

To help troubleshoot performance issues, you should enable parse avoidance debug logging. To do so, include a capital letter “P” in the argument of the --emake-debug=<arguments> option. CloudBees engineering and support staff use the eMake debug logs to help troubleshoot problems. For more information about debug logging and log levels, see the “ eMake Debug Log Levels.

Debug Log Example

You can look in "P" debug logging for an explanation of why a cached result was found to be obsolete:

WK01: 0.062173 Input changed: job:J08345218 slot:b6189634a793fee2fe1929fbf47cc4e4 path:/home/aeolus/t/Makefile thenSize:175 nowSize:176 ignore:0 WK01: 0.062228 Input changed: job:J08345218 slot:b6189634a793fee2fe1929fbf47cc4e4 path:/home/aeolus/t/fog.mk thenSize:0 nowSize:1 ignore:0 ... WK01: 0.062284 Cache slot is obsolete: job:J08345218 slot:b6189634a793fee2fe1929fbf47cc4e4

Notice that " Makefile " and " fog.mk " were both bigger; either of those changes would have triggered a reparse. The "ignore:0" comments indicate that --emake-parse-avoidance-ignore-path was not used to ignore the changes.

Using Key Files for Debugging

To discover why a new cache slot was used, look in "P" debug logging for the name of the "key" file for that new cache slot; for example:

WK01: 11.853035 Saved slot definition: .emake/cache.11/i686_Linux/d9/0f/79/4fb975e8ff94dfa2569a303e62.new.2NPR8J/key
The ".new.2NPR8J" portion of the slot directory name should have already been removed automatically by eMake before you need to access the key file.

Then diff that key file against other key files in sibling directories to see what parameters changed. To determine which slot directories to compare, grep for the parse job IDs in "P" debug logging. You can get parse job IDs from annotation using Insight. Each key file contains an artificial cd command to express the working directory, all non-ignored environment variables, and the non-ignored command-line arguments. Ignored command-line arguments and environment variables are omitted from key files. The key file is stored in a directory whose path name is derived from the md5sum of the file itself. Also, the command line might have been normalized to one that is equivalent to the original one. eMake quotes special characters according to UNIX, Linux, and Cygwin sh shell rules.

Ignored emake Command-Line Options

The following emake command-line options do not affect which cache slot is chosen:

--emake-annodetail --emake-annofile --emake-annoupload --emake-assetdir --emake-big-file-size --emake-build-label --emake-class --emake-cluster-timeout --emake-cm --emake-debug --emake-hide-warning --emake-history --emake-history-force --emake-historyfile --emake-idle-time
--emake-job-limit --emake-ledger --emake-ledgerfile --emake-localagents --emake-logfile --emake-logfile-mode --emake-maxagents --emake-maxlocalagentsREMOVED IN 9.1 AS PER https://cloudbees.atlassian.net/browse/EC-12154. --emake-mem-limit --emake-parse-avoidance-ignore-path --emake-pedantic --emake-priority --emake-readdir-conflicts --emake-resource --emake-showinfo --emake-tmpdir --emake-yield-localagentsREMOVED IN 9.1 AS PER https://cloudbees.atlassian.net/browse/EC-12154.

Ignored Environment Variables

By default, the following environment variables do not affect which cache slot is chosen. You can specify additional environment variables with --emake-parse-avoidance-ignore-env. You must use --emake-parse-avoidance-ignore-env once for each variable to ignore.

_ (underscore) ANDROID_BUILD_TOP* ANDROID_HOST_OUT* BUILD_DISPLAY_NAME BUILD_ID BUILD_NUMBER BUILD_TIMESTAMP BUILD_TAG CI_JENKINS_BUILD_ID CI_JENKINS_BUILD_NUMBER CI_JENKINS_BUILD_TAG CI_JENKINS_HUDSON_SERVER_COOKIE CI_JENKINS_JENKINS_SERVER_COOKIE COLORFGBG COLUMNS COMMANDER_HOME COMMANDER_HTTPS_PORT COMMANDER_JOBID COMMANDER_JOBSTEPID COMMANDER_PLUGINS COMMANDER_PORT COMMANDER_RESOURCENAME COMMANDER_SERVER COMMANDER_SESSIONID COMMANDER_WORKSPACE COMMANDER_WORKSPACE_NAME COMMANDER_WORKSPACE_UNIX COMMANDER_WORKSPACE_WINDRIVE COMMANDER_WORKSPACE_WINUNC DBUS_SESSION_BUS_ADDRESS DESKTOP_SESSION DISPLAY ECLOUD_AGENT_NUM ECLOUD_BUILD_CLASS ECLOUD_BUILD_COUNTER ECLOUD_BUILD_ID ECLOUD_BUILD_TAG ECLOUD_BUILD_TYPE ECLOUD_RECURSIVE_COMMAND_FILE EMAKE_APP_VERSION EMAKE_BUILD_MODE EMAKE_DEBUG EMAKE_LEDGER_CONTEXT EMAKE_LICENSE_FILE EMAKE_MAKE_ID EMAKE_PARSEFILE EMAKE_RLOGFILE EMAKE_ROOT GPG_AGENT_INFO HOSTNAME HUDSON_COOKIE HUDSON_HOME HUDSON_SERVER_COOKIE HUDSON_URL JACK_CLIENT_SETTING* JENKINS_COOKIE JENKINS_HOME
JENKINS_SERVER_COOKIE JENKINS_URL JOB_BASE_NAME JOB_DISPLAY_URL JOB_NAME KDE_FULL_SESSION KDE_MULTIHEAD KDE_SESSION_UID KDE_SESSION_VERSION KONSOLE_DBUS_SERVICE KONSOLE_DBUS_SESSION LINES LS_COLORS OLDPWD PROMPT_COMMAND PS1 RUN_CHANGES_DISPLAY_URL RUN_DISPLAY_URL SESSION_MANAGER SHELL_SESSION_ID SHLVL SSH_AGENT_PID SSH_AUTH_SOCK SSH_CLIENT SSH_CONNECTION SSH_TTY TEMP TERM TERMCAP TMP TMPDIR WINDOWID WINDOWPATH WRAPPER_ARCH WRAPPER_BIN_DIR WRAPPER_BITS WRAPPER_CONF_DIR WRAPPER_DESCRIPTION WRAPPER_DISPLAYNAME WRAPPER_EVENT_NAME WRAPPER_EVENT_WRAPPER_PID WRAPPER_FILE_SEPARATOR WRAPPER_HOSTNAME WRAPPER_INIT_DIR WRAPPER_JAVA_HOME WRAPPER_EVENT_JVM_ID WRAPPER_EVENT_JVM_PID WRAPPER_LANG WRAPPER_NAME WRAPPER_OS WRAPPER_PATH_SEPARATOR WRAPPER_PID WRAPPER_SYSMEM_PP.P WRAPPER_WORKING_DIR XCURSOR_THEME XDG_SESSION_COOKIE XDM_MANAGED
BUILD_DISPLAY_NAME BUILD_ID BUILD_NUMBER BUILD_TAG CI_JENKINS_BUILD_ID CI_JENKINS_BUILD_NUMBER CI_JENKINS_BUILD_TAG CI_JENKINS_HUDSON_SERVER_COOKIE CI_JENKINS_JENKINS_SERVER_COOKIE COMMANDER_JOBID COMMANDER_JOBSTEPID COMMANDER_SESSIONID COMMANDER_WORKSPACE COMMANDER_WORKSPACE_UNIX COMMANDER_WORKSPACE_WINDRIVE COMMANDER_WORKSPACE_WINUNC COMMANDER_RESOURCENAME EMAKE_PARSEFILE EMAKE_RLOGFILE ECLOUD_AGENT_NUM ECLOUD_BUILD_CLASS ECLOUD_BUILD_COUNTER ECLOUD_BUILD_ID ECLOUD_BUILD_TAG ECLOUD_BUILD_TYPE EMAKE_APP_VERSION EMAKE_BUILD_MODE ECLOUD_RECURSIVE_COMMAND_FILE EMAKE_LEDGER_CONTEXT EMAKE_MAKE_ID TMP TEMP TMPDIR

_ HOSTNAME OLDPWD SHLVL SSH_AGENT_PID SSH_AUTH_SOCK SSH_CLIENT SSH_CONNECTION SSH_TTY GPG_AGENT_INFO COLUMNS LINES TERM TERMCAP COLORFGBG LS_COLORS DISPLAY DESKTOP_SESSION SHELL_SESSION_ID SESSION_MANAGER DBUS_SESSION_BUS_ADDRESS KDE_FULL_SESSION KDE_MULTIHEAD KDE_SESSION_UID KDE_SESSION_VERSION KONSOLE_DBUS_SERVICE KONSOLE_DBUS_SESSION WINDOWID WINDOWPATH XCURSOR_THEME XDG_SESSION_COOKIE XDM_MANAGED
EMAKE_PARSEFILE EMAKE_RLOGFILE ECLOUD_AGENT_NUM ECLOUD_BUILD_CLASS ECLOUD_BUILD_COUNTER ECLOUD_BUILD_ID ECLOUD_BUILD_TAG ECLOUD_BUILD_TYPE EMAKE_APP_VERSION EMAKE_BUILD_MODE ECLOUD_RECURSIVE_COMMAND_FILE EMAKE_LEDGER_CONTEXT EMAKE_MAKE_ID TMP TEMP TMPDIR _ OLDPWD SHLVL SSH_AGENT_PID SSH_AUTH_SOCK SSH_CLIENT SSH_CONNECTION SSH_TTY
EMAKE_PARSEFILE
EMAKE_RLOGFILE
ECLOUD_AGENT_NUM
ECLOUD_BUILD_CLASS
ECLOUD_BUILD_COUNTER
ECLOUD_BUILD_ID
ECLOUD_BUILD_TAG
ECLOUD_BUILD_TYPE
EMAKE_APP_VERSION
EMAKE_BUILD_MODE
ECLOUD_RECURSIVE_COMMAND_FILE
WRAPPER_ARCH
WRAPPER_BIN_DIR
WRAPPER_BITS
WRAPPER_CONF_DIR
WRAPPER_FILE_SEPARATOR
WRAPPER_HOSTNAME
WRAPPER_INIT_DIR
WRAPPER_JAVA_HOME
WRAPPER_LANG
WRAPPER_OS
WRAPPER_PATH_SEPARATOR
GPG_AGENT_INFO COLUMNS LINES TERM TERMCAP COLORFGBG LS_COLORS DISPLAY DESKTOP_SESSION SHELL_SESSION_ID SESSION_MANAGER DBUS_SESSION_BUS_ADDRESS KDE_FULL_SESSION KDE_MULTIHEAD KDE_SESSION_UID KDE_SESSION_VERSION KONSOLE_DBUS_SERVICE KONSOLE_DBUS_SESSION WINDOWID WINDOWPATH XCURSOR_THEME XDG_SESSION_COOKIE XDM_MANAGED
WRAPPER_PID
WRAPPER_WORKING_DIR
WRAPPER_SYSMEM_PP.P
WRAPPER_NAME
WRAPPER_DISPLAYNAME
WRAPPER_DESCRIPTION
WRAPPER_EVENT_JVM_ID
WRAPPER_EVENT_JVM_PID
WRAPPER_EVENT_NAME
WRAPPER_EVENT_WRAPPER_PID

Ignored Android Environment Variables

The following default set of environment variables do not affect which cache slot is chosen. This list is included as a default set only when you use --emake-android-version (otherwise, you must specify them explicitly).

ANDROID_BUILD_TOP ANDROID_HOST_OUT
JACK_CLIENT_SETTING