Resume building releases with Xcode 13 #7426
Conversation
… 3 is broken" This reverts commit cef5965.
We already did this for non-SPM builds as it makes the build ~20% faster and write about half as much data.
This has been broken ever since we started hiding the project file from Carthage because sed can't do in-place edits on a symlink (but ex can).
| let str: String | ||
| switch self { | ||
| case let .decimal128(d): | ||
| return d | ||
| case let .int64(i): | ||
| return try? Decimal128(string: String(i)) | ||
| str = String(i) | ||
| case let .int32(i): | ||
| return try? Decimal128(string: String(i)) | ||
| str = String(i) | ||
| case let .double(d): | ||
| return try? Decimal128(string: String(d)) | ||
| str = String(d) | ||
| default: | ||
| return nil | ||
| } | ||
| return try? Decimal128(string: str) |
There was a problem hiding this comment.
This change made this function compile in 5ms rather than 150ms (as reported by -debug-time-function-bodies)
| public func distance(to other: Decimal128) -> Stride { | ||
| other - self | ||
| public func distance(to other: Decimal128) -> Decimal128 { | ||
| unsafeDowncast(other.decimalNumber(bySubtracting: self), to: Decimal128.self) |
There was a problem hiding this comment.
Overloaded operators take a really long time to type-check so each of these were taking close to 100ms to build and now are <1ms. unsafeDowncast() ended up not speeding up the build but made the release binary a bit smaller (but not enough to justify listing it in the changelog).
There was a problem hiding this comment.
Is it worth adding some comments to say why you use unsafeDowncast? Its not initially apparent that it improves binary size.
There was a problem hiding this comment.
But I suppose you could just always trace the git blame
There was a problem hiding this comment.
The difference between unsafeDowncast() and as! is that the latter inserts runtime checks to verify that it actually has the type, while unsafeDowncast() doesn't in optimized builds. Those checks ended up being a significant percentage of this file due to how little the Swift wrapper actually does, but in absolute terms they weren't large.
| @@ -165,7 +165,7 @@ public extension ObjectiveCSupport { | |||
| return convertBson(object: value) | |||
| } | |||
| return .null | |||
| }.map { $0 == .null ? nil : $0 }) | |||
| }.map { (v: AnyBSON) -> AnyBSON? in v == .null ? nil : v }) | |||
There was a problem hiding this comment.
Similarly this type hint made this function a lot faster to build.
| @@ -649,8 +649,7 @@ case "$COMMAND" in | |||
| export ASAN_OPTIONS='check_initialization_order=true:detect_stack_use_after_return=true' | |||
| fi | |||
| xcrun swift package resolve | |||
| find .build -name views.cpp -delete | |||
| xcrun swift test --configuration "$(echo "$CONFIGURATION" | tr "[:upper:]" "[:lower:]")" $SANITIZER | |||
| xcrun swift test -Xcc -g0 --configuration "$(echo "$CONFIGURATION" | tr "[:upper:]" "[:lower:]")" $SANITIZER | |||
There was a problem hiding this comment.
The non-SPM pull request jobs were already disabling debug symbols so this brings SPM in line with them. This knocks ~20% off the build time and cuts the total size of all the object files and such in half, which might fix some of the jobs which were timing out.
| @@ -106,7 +106,7 @@ xctest() { | |||
| ) | |||
| elif [[ $NAME == SwiftPackageManager* ]]; then | |||
| if [ -n "$sha" ]; then | |||
| sed -i '' 's@branch = "master"@branch = "'"$sha"'"@' "$DIRECTORY/$NAME.xcodeproj/project.pbxproj" | |||
| ex '+%s@branch = "master"@branch = "'"$sha"'"@' -scwq "$DIRECTORY/$NAME.xcodeproj/project.pbxproj" | |||
There was a problem hiding this comment.
sed -i doesn't work on symlinks and this file has been a symlink for a while. ex (the line-editor part of vim) does and is a reasonably drop-in replacement.
| @@ -133,7 +133,9 @@ xctest() { | |||
| local scheme=(-scheme "$NAME") | |||
|
|
|||
| # Ensure that dynamic framework tests try to use the correct version of the prebuilt libraries. | |||
| sed -i '' 's@/realm-swift-latest@/realm-swift-latest/'"${REALM_XCODE_VERSION}"'@' "$DIRECTORY/$NAME.xcodeproj/project.pbxproj" | |||
| if grep '/realm-swift-latest' "$DIRECTORY/$NAME.xcodeproj/project.pbxproj"; then | |||
There was a problem hiding this comment.
Unlike sed, ex reports failure if the expression isn't present in the file so check first.
tgoyne
requested review from
leemaguire and
jsflax
and removed request for
leemaguire
Yes, and the main thing is showed was that the Swift frontend isn't a meaningful portion of our build time and there isn't much room for improvement there. Everything in non-test code left that's >10ms adds up to a total of ~300ms and most of that is in the long tail of 12ms functions. None of the test code appears to be hitting any particularly bad cases for the compiler either (at least with Xcode 13). |
#7401 worked around the problem so we want to start packaging it again. I also fixed the remaining things making cocoa-pipeline fail now that the cocoapods release required for that is out, and this branch built successfully: https://ci.realm.io/blue/organizations/jenkins/cocoa-pipeline/detail/cocoa-pipeline/1497/pipeline/