XCRemoteCache is a remote cache tool for Xcode projects. It reuses target artifacts generated on a remote machine, served from a simple REST server.

• How and Why?
  • Accurate target input files
    • New file added to the target
  • Debug symbols
  • Performance optimizations
  • Focused targets
• How to integrate XCRemoteCache with your Xcode project?
  • 1. Download XCRemoteCache
  • A. Automatic integration
    • 2. Create a minimal XCRemoteCache configuration
    • 3. Run automatic integration script
      • 3a. Producer side
      • 3b. Consumer side
        • A full list of xcprepare integrate supported options
  • B. Manual integration
    • 2. Configure XCRemoteCache
    • 3. Call xcprepare
    • 4. Integrate with the Xcode project
    • 5. Configure LLDB source-map (Optional)
    • 6. Producer mode - Artifacts generation
      • 6a. Configure producer mode
      • 6b. Fill the cache
      • 6c. Mark commit sha
• A full list of configuration parameters:
• Backend cache server
  • Sample REST cache server from a docker image
  • Amazon S3 and Google Cloud Storage
• CocoaPods plugin
• Requirements
• Apple silicon support
  • Artifacts per architecture (Recommended)
  • Fat artifacts
• Limitations
• FAQ
• Development
• Architectural Designs
• Release
  • Releasing CocoaPods plugin
  • Building release package
• Contributing
• Code of conduct
• License
• Security Issues?

The caching mechanism is based on remote artifacts that should be generated and uploaded to the cache server for each commit on a master branch, preferably as a part of CI/CD step. Xcode products are not portable between different Xcode versions, each XCRemoteCache artifact is linked with a specific Xcode build number that generated it. To support multiple Xcode versions, artifacts generation should happen for each Xcode version.

The artifact reuse flow is as follows: XCRemoteCache performs a target precheck (aka prebuild) and if a fingerprint for local sources matches the one computed on a generation side, several compilation steps wrappers (e.g. xcswiftc, xccc, xclibtool) mock corresponding compilation step(s) and linking (or archiving) moves the cached build artifact to the expected location.

Multiple commits that have the same target sources reuse artifact package on a remote server.

Finding a precise list of input and dependency files is a non-trivial task as Xcode heavily relies on implicit target dependencies. It means that Xcode is trying to use required dependencies from provided search paths and looks up DerivedData's product dir. To find a narrow list of files to compute fingerprint hash, XCRemoteCache fetches a meta.json file from a server which contains remotelly generated fingerprint hash and a list of input files that should constitute a fingerprint. That list of input files was produced during the artifact generation process, based on .d output from clang and swift compilers.

Before building a project in Xcode, XCRemoteCache needs to find the best git commit sha for which artifacts will be used. This happens as a part of the xcprepare execution, which should be called after each merge or switching a branch. xcprepare finds a list of 10 most recent common sha with the remote repo branch using git's first-parent strategy and selects the newest one for which all artifacts have been uploaded.

The generation side is responsible to call xcprepare mark subcommand after each successful build. Marking process creates an empty marker file on a remote cache server with a given format: #{commmmitSha}-#{TargetName}-#{Configuration}-#{Platform}-#{XcodeBuildNumber}-#{ContextBuildSettings}-#{SchemaID}.json.

xcprepare makes HEAD requests for all identified shas, picks the newest one for which a marker file exists remotely, and saves it in the text form to the arc.rc file. That file informs the prebuild phase which meta file should be fetched to get a list of target dependency files.

Considering in the hash fingerprint only a list of previously observed files can give invalid results if a build contains a new source file as it isn't considered in the hash.

For a new .swift file in a swift-only target, xcswiftc automatically recognizes that case and forces local compilation of the entire target. For Objective-C or mixed targets, fallbacking to the local compilation is more difficult as some previous invocations (either xccc or xcswiftc) could already be finished with no-operation. To mitigate that, each wrapper appends invocation call to a side file (history.compile) just in case some other process would need to compile the entire target locally. If that happens, compilation of the newly added file acquires a target-wide lock that stops other wrapper invocations, executes already mocked steps one by one to backfill already skipped compilation steps.

Binaries built with "debug symbols: enabled" embed source file absolute paths so compilation products cannot be directly ported between two machines with different source roots. Otherwise, LLDB debugger is not able to correlate a set of currently executing machine instructions with a local file that produced it. To mitigate that, XCRemoteCache recommends adding a custom C and Swift debug flags prefix-map for all XCRemoteCache builds. These flags ensure that all binaries, generated locally and downloaded from a remote server, have the same debug symbols absolute paths which are translated to an actual local path at the beginning of the LLDB session.

XCRemoteCache involves several optimization techniques:

• Local HTTP cache stores all responses from the remote server at ~/Library/Caches/XCRemoteCache/
• Prebuild and postbuild steps leverage Xcode's discovered dependency file to avoid recomputing fingerprint hashes if none of the input files has changed
• A wrapper for the clang compilation is a C program, generated and compiled during the xcprepare step. It is called many times to compile each *.(m|c) file and accessing a disk to read a configuration would introduce a significant slowdown, especially if a project contains a lot of Objective-C files. As a remedy, xcprepare reads the XCRemoteCache configuration only once and embeds all configurable fields directly into the xccc binary
• arc.rc, generated by xcprepare, gets file modification equal to the commit date if refers. arc.rc is included in the discovered dependency file, touching it in the xcprepare would automatically invalidate previous XCRemoteCache prebuild step and force redundant fingerprint checks. By syncing the mdate with a git commit, Xcode avoids prebuild steps unless the remote cache commit has changed
• If a target cache miss happens, XCRemoteCache disables cache for that target until a commit sha in arc.rc changes. That bypasses a fingerprint computation for incremental builds

If a list of targets that can have dirty sources is limited, XCRemoteCache can be configured with focused targets, specified in .rcinfo.

By default, all targets are focused and these compare local fingerprint with one available remotely and fallbacks to the local compilation if it doesn't match. Non-focused targets, called 'thin' targets, always use cached artifacts what eliminates a fingerprint computation. Thin targets should contain only a single compilation file with thin_target_mock_filename, e.g. standin.swift or standin.m.

To enable XCRemoteCache in the existing .xcodeproj you need to add extra build settings and build phases to targets that you want to cache.

You can do that in an automatic way, using the XCRemoteCache-provided integration command, or manually modify your Xcode project.

From the Github Releases page, download the XCRemoteCache bundle zip. Unzip the bundle to a directory next to your .xcodeproj.

The following steps will assume the bundle has been unzipped to xcremotecache dir, placed next to the .xcodeproj.

Create .rcinfo yaml file next to the .xcodeproj with a minimum set of configuration entries, like:

primary_repo: https://yourRepo.git
cache_addresses:
- https://xcremotecacheserver.com

Execute a command that modifies <yourProject.xcodeproj>:

xcremotecache/xcprepare integrate --input <yourProject.xcodeproj> --mode producer --final-producer-target <YourMainTarget>

Execute a command that modifies <yourProject.xcodeproj>:

xcremotecache/xcprepare integrate --input <yourProject.xcodeproj> --mode consumer

Create yaml configuration file .rcinfo, next to the .xcodeproj, with your full XCRemoteCache configuration, according to the parameters list e.g.:

primary_repo: https://yourRepo.git
cache_addresses:
 - https://xcremotecacheserver.com
repo_root: "."
remote_commit_file: arc.rc
xccc_file: xcremotecache/xccc

Execute xcprepare --configuration #Configuration# --platform #platform# command after each merge or rebase with the primary branch. Otherwise, the remote cache artifacts may be outdated and final hit rate may be poor.

The xcprepare application saves arc.rc file on a disk and prints a summary to the standard output. The printed recommended_remote_address is just a recommendation which cache remote server use. It is up to the integration tooling to decide if it makes sense. If so, the project's .rcinfo should define that value as recommended_remote_address parameter.

Example:

$ xcremotecache/xcprepare --configuration Debug --platform iphonesimulator
result: true
commit: aabbccc00
age: 0
recommended_remote_address: https://xcremotecacheserver.com

Configure Xcode targets that should use XCRemoteCache:

1. Override Build Settings:

• CC - xccc_file from your .rcinfo configuration (e.g. xcremotecache/xccc)
• SWIFT_EXEC - location of xcprepare (e.g. xcremotecache/xcswiftc)
• LIBTOOL - location of xclibtool (e.g. xcremotecache/xclibtool)
• LIPO - location of xclipo (e.g. xcremotecache/xclipo)
• LD - location of xcld (e.g. xcremotecache/xcld)
• LDPLUSPLUS - location of xcldplusplus (e.g. xcremotecache/xcldplusplus)
• XCRC_PLATFORM_PREFERRED_ARCH - $(LINK_FILE_LIST_$(CURRENT_VARIANT)_$(PLATFORM_PREFERRED_ARCH):dir:standardizepath:file:default=arm64)
• SWIFT_USE_INTEGRATED_DRIVER - NO (required in Xcode 14.0+)

2. Add a Prebuild build phase (before compilation):

• command: "$SCRIPT_INPUT_FILE_0"
• input files: location of xcprebuild (e.g. xcremotecache/xcprebuild)
• output files:
  • $(TARGET_TEMP_DIR)/rc.enabled
  • $(DWARF_DSYM_FOLDER_PATH)/$(DWARF_DSYM_FILE_NAME)
• discovery dependency file: $(TARGET_TEMP_DIR)/prebuild.d

3. Add Postbuild build phase (after compilation):

• command: "$SCRIPT_INPUT_FILE_0"
• input files: location of xcpostbuild command (e.g. xcremotecache/xcpostbuild)
• output files:
  • $(TARGET_BUILD_DIR)/$(MODULES_FOLDER_PATH)/$(PRODUCT_MODULE_NAME).swiftmodule/$(XCRC_PLATFORM_PREFERRED_ARCH).swiftmodule.md5
  • $(TARGET_BUILD_DIR)/$(MODULES_FOLDER_PATH)/$(PRODUCT_MODULE_NAME).swiftmodule/$(XCRC_PLATFORM_PREFERRED_ARCH)-$(LLVM_TARGET_TRIPLE_VENDOR)-$(SWIFT_PLATFORM_TARGET_PREFIX)$(LLVM_TARGET_TRIPLE_SUFFIX).swiftmodule.md5
• discovery dependency file: $(TARGET_TEMP_DIR)/postbuild.d

Rewriting source-map is required to support debugging and hit breakpoints, see Debug symbols.

1. Ooverride the following Build Settings for all targets:

• XCRC_SRCROOT - /xxxxxxxxxx (or any other arbitrary string for your project)
• add -debug-prefix-map $(SRCROOT)=$(XCRC_SRCROOT) to OTHER_SWIFT_FLAGS. If it doesn't exist, define it as $(inherited) -debug-prefix-map $(SRCROOT)=$(XCRC_SRCROOT)
• add -fdebug-prefix-map=$(SRCROOT)=$(XCRC_SRCROOT) to OTHER_CFLAGS. If it doesn't exists, define it as $(inherited) -fdebug-prefix-map=$(SRCROOT)=$(XCRC_SRCROOT)

2. Add settings set target.source-map /xxxxxxxxxx /Users/account/src/PathToTheProject to ~/.lldbinit on end machine that builds a project with XCRemoteCache

XCRC_SRCROOT arbitrary path should be project-exclusive to avoid clashing.

Tip: In some rare cases, Xcode caches ~/.lldbinit content so make sure to restart Xcode after the modification.

XCRemoteCache can operate in two main modes: consumer (default) tries to reuse artifacts available on the remote server and producer is used to generate all artifacts - it builds all targets locally and uploads meta and artifact files to the remote cache server.

To enable the producer mode, configure it directly in the .rcinfo file.

Optionally, you can define extra_configuration_file in a .rcinfo with a path to the other yaml file that will override the default configuration in .rcinfo. That approach can be useful if you want to track main .rcinfo and keep your local configuration out of git.

Build the project from Xcode or using xcodebuild

Once all artifacts have been uploaded, "mark a build" using xcprepare mark command:

$ xcremotecache/xcprepare mark --configuration Debug --platform iphonesimulator

That command creates an empty file on a remote server which informs that for given sha, configuration, platform, Xcode versions etc. all artifacts are available.

Note that for the producer mode, the prebuild build phase and xccc, xcld, xcldplusplus, xclibtool, xclipo wrappers become no-op, so it is recommended to not add them for the producer mode.

If a static library target contains a mixed target with a bridging header exposing an enum from ObjC in a public Swift API, your custom script that moves *-Swift.h to the shared location, it should also move *-Swift.h.md5 next to it.

Example:

ditto "${SCRIPT_INPUT_FILE_0}" "${SCRIPT_OUTPUT_FILE_0}"

where

• SCRIPT_INPUT_FILE_0="$(DERIVED_SOURCES_DIR)/$(SWIFT_OBJC_INTERFACE_HEADER_NAME)"
• SCRIPT_OUTPUT_FILE_0="$(BUILT_PRODUCTS_DIR)/include/$(PRODUCT_MODULE_NAME)/$(SWIFT_OBJC_INTERFACE_HEADER_NAME)"

ditto "${SCRIPT_INPUT_FILE_0}" "${SCRIPT_OUTPUT_FILE_0}"
[ -f "${SCRIPT_INPUT_FILE_1}" ] && ditto "${SCRIPT_INPUT_FILE_1}" "${SCRIPT_OUTPUT_FILE_1}" || rm -f "${SCRIPT_OUTPUT_FILE_1}"

where

• SCRIPT_INPUT_FILE_0="$(DERIVED_SOURCES_DIR)/$(SWIFT_OBJC_INTERFACE_HEADER_NAME)"
• SCRIPT_INPUT_FILE_1="$(DERIVED_SOURCES_DIR)/$(SWIFT_OBJC_INTERFACE_HEADER_NAME).md5"
• SCRIPT_OUTPUT_FILE_0="$(BUILT_PRODUCTS_DIR)/include/$(PRODUCT_MODULE_NAME)/$(SWIFT_OBJC_INTERFACE_HEADER_NAME)"
• SCRIPT_OUTPUT_FILE_1="$(BUILT_PRODUCTS_DIR)/include/$(PRODUCT_MODULE_NAME)/$(SWIFT_OBJC_INTERFACE_HEADER_NAME).md5"

Note: This step is not required if at least one of these is true:

• you build a framework (not a static library)
• you don't expose NS_ENUM type from ObjC to Swift via a bridging header

As a cache server, XCRemoteCache may use any REST server that supports PUT, GET and HEAD methods.

For the development phase, you can try the simplest cache server available as a docker image in backend-example. For the production environment, it is recommended to configure a reliable, fast server, preferrably located in a close proximity to developer's machines.

Out-of-the-box, XCRemoteCache supports V4 Signature Authorization used by Amazon's S3 and Google's GCS. Altenatively, if your server has a customized authentication procedure, you can add extra HTTP request headers with request_custom_headers configuration property.

To run a local instance of a server, use a snippet which exposes a cache endpoint under http://localhost:8080/cache:

docker build -t xcremotecache-demo-server backend-example
docker run -it --rm -d -p 8080:8080 --name xcremotecache xcremotecache-demo-server

As the docker image saves all files in a container non-persistent storage, to reset cache's content, just restart it:

# stop the container
docker kill xcremotecache
# run a new instance of the image
docker run -it --rm -d -p 8080:8080 --name xcremotecache xcremotecache-demo-server

To review all files stored in the cache server, navigate to the container's cache root directory:

docker exec -w /tmp/cache -it xcremotecache /bin/bash

XCRemoteCache supports Amazon S3 and Google Cloud Storage buckets to be used as cache servers using the Amazon v4 Signature Authorization.

To set it up use the configuration parameters aws_secret_key, aws_access_key, aws_region, and aws_service in the .rcinfo file. Specify the URL to the bucket in cache-addresses field in the same file.

XCRemoteCache also supports AWS Temporary Access Keys. Use additional aws_security_token parameter combined with aws_secret_key, aws_access_key to set it up. This page describes how to receive a security token.

Example

...
cache_addresses:
 - https://bucketname.s3.eu-central-1.amazonaws.com/
aws_secret_key: <SECRET_KEY>
aws_access_key: <ACCESS_KEY>
aws_region: eu-central-1
aws_service: s3
...

Retention Policy: Buckets usually have a retention policy option which ensures objects are retained for a certain amount of time and won't be modified or deleted. Keep this option short or disable it to avoid errors in case multiple builds are done consecutively on the producer side for the same configuration.

Head over to our cocoapods-plugin docs to see how to integrate XCRemoteCache in your CocoaPods project.

If all of your machines (both producer and all consumers have the same architecture, either Intel or Apple Silicon), you don't have to do anything.

XCRemoteCache supports building artifacts for Apple silicon consumers. Is it recommended to build separately for x86_64 and arm64 architectures to have single-architecture artifacts that do not require downloading irrelevant binaries. Here are required steps if you want to support both Intel and Apple silicon consumers.

• Building for a simulator on a producer: run a first build for x86_64, clean a build and build again for arm64, e.g.:

xcodebuild ARCHS=x86_64 ONLY_ACTIVE_ARCH=NO build ...
xcodebuild clean
xcodebuild ARCHS=arm64 ONLY_ACTIVE_ARCH=NO build ...

If you prefer to generate far artifacts (with both Intel and Apple silicon binaries), you can disable "Build Archive Architecture Only" on a producer side, e.g.

xcodebuild ONLY_ACTIVE_ARCH=NO build ...

Note: This setup is not recommended and may not be supported in future XCRemoteCache releases.

• The repo under git version control
• Xcode 11.4+
• Xcode New Build System
• Current Xcode location set by xcode-select
• Using the default Xcode Toolchain
• Recommended: multi-targets Xcode project
• Recommended: do not use fast-forward PR strategy (use merge or squash instead)
• Recommended: avoid DWARF with dSYM File "Debug Information Format" build setting. Use DWARF instead
• Recommended: avoid having a symbolic link in the source root (e.g. placing a project in /tmp)

• Swift Package Manager (SPM) dependencies are not supported. Because SPM does not allow customizing Build Settings, XCRemoteCache cannot specify clang and swiftc wrappers that control if the local compilation should be skipped (cache hit) or not (cache miss)
• Filenames with _vers.c suffix are reserved and cannot be used as a source file
• All compilation files should be referenced via the git repo root. Referencing /AbsolutePath/someOther.swift or ../../someOther.swift that resolve to the location outside of the git repo root is prohibited.
• The new Swift driver (introduced by default in Xcode 14.0) is not supported and has to be disabled when using XCRemoteCache

Follow the FAQ page.

Follow the Development guide. It has all the information on how to get started.

Follow the Architectural designs document that describes and documents XCRemoteCache designs and implementation details.

To release a version, in Releases draft a new release with v0.3.0{-rc0} tag format. Packages with binaries will be automatically uploaded to the GitHub Releases page.

Bump a gem version defined in gem_version.rb and create a new release described above.

A plugin is automatically uploaded to RubyGems if a given version doesn't exist yet.

To build a release zip package for a single platform (e.g. x86_64-apple-macosx, arm64-apple-macosx), call:

rake 'build[release, x86_64-apple-macosx]'

The zip package will be generated at releases/XCRemoteCache.zip.

Create a new issue with as many details as possible.

Reach us at the #xcremotecache channel in Slack.

We feel that a welcoming community is important and we ask that you follow Spotify's Open Source Code of Conduct in all interactions with the community.

This project adheres to the Open Code of Conduct. By participating, you are expected to honor this code.

Copyright 2021 Spotify AB

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Please report sensitive security issues via Spotify's bug-bounty program (https://hackerone.com/spotify) rather than GitHub.