Fix multi-arch cross-compiling #13108

zbowling · 2017-11-28T04:22:48Z

On platforms without fat binaries (anything but Darwin), we now build
and install stdlibs for each arch in it's own folder. Namely
lib/swift/<platform>/<arch> instead of lib/swift/<platform>.

(Split from #12955)

zbowling · 2017-11-28T04:35:46Z

cmake/modules/AddSwift.cmake

@@ -327,7 +327,7 @@ function(_add_variant_link_flags)
    ${ARGN})

  precondition(LFLAGS_SDK MESSAGE "Should specify an SDK")
-  precondition(LFLAGS_SDK MESSAGE "Should specify an architecture")
+  precondition(LFLAGS_ARCH MESSAGE "Should specify an architecture")


Fixed an unrelated copy/paste bug with this arg.

zbowling · 2017-11-28T04:37:38Z

cmake/modules/AddSwift.cmake

-      "${SWIFT_NATIVE_SWIFT_TOOLS_PATH}/../lib/swift/${SWIFTLIB_SINGLE_SUBDIR}"
+      "${SWIFT_NATIVE_SWIFT_TOOLS_PATH}/../lib/swift/${SWIFTLIB_SINGLE_SUBDIR}")
+
+  if("${SWIFT_SDK_${SWIFTLIB_SINGLE_SDK}_OBJECT_FORMAT}" STREQUAL "ELF")


Not sure if I should gate on elf or gate on mach. @compnerd windows?

PE/COFF doesn't have fat binaries either. I think that we should make this a Mach check.

Lines 1650 and 1743 use an IS_DARWIN check instead.

Yeah, I can switch to that.

jrose-apple

Driver parts look okay to me. I'll let Ross and others comment on the CMake.

jrose-apple · 2017-12-01T01:06:01Z

lib/Driver/ToolChains.cpp

@@ -1016,6 +1016,17 @@ static void getRuntimeLibraryPath(SmallVectorImpl<char> &runtimeLibPath,
                          getPlatformNameForTriple(TC.getTriple()));
 }

+/// On Darwin we link fat libs in the root of the library path and on other OS
+/// targets we look up libraries in the target arch's subfolder.


Nitpick: This comment doesn't really describe the function.

Ah true. What I was thinking when I changed where we used it and not what the function does.

gparker42 · 2017-12-01T01:12:32Z

cmake/modules/AddSwift.cmake

-          # Note this thin library.
-          list(APPEND THIN_INPUT_TARGETS ${VARIANT_NAME})
+            if (SWIFTLIB_IS_STDLIB AND SWIFTLIB_STATIC)
+              # Add dependencies on the (not-yet-created) custom lipo target.


This comment can't be right. This is inside the non-Darwin no-lipo case.

Ah good catch. Copy-pasta.

compnerd · 2017-12-03T18:36:33Z

I will try to take a closer look at this soon. But, we should try to get this patch taken care of soon. There is at least one patch from me (#12480) and one from @troughton (#13140) which is trying to address the problems caused by the fat vs thin binary handling in the build.

zbowling · 2017-12-03T23:29:30Z

@compnerd Yeah, I see what you are doing in that other patch. I'll try to fix this one up really quick.

zbowling · 2017-12-03T23:55:11Z

Fixed issues. can someone kick off the CI bot for me? I can't test locally this weekend.

phausler · 2017-12-04T00:07:04Z

@swift-ci please test

swift-ci · 2017-12-04T00:07:28Z

Build failed
Swift Test OS X Platform
Git Sha - ce53ad3f70deb1c26d9929f7cb1f66e875454209

swift-ci · 2017-12-04T00:08:27Z

Build failed
Swift Test Linux Platform
Git Sha - ce53ad3f70deb1c26d9929f7cb1f66e875454209

phausler · 2017-12-04T00:16:14Z

@swift-ci please test

swift-ci · 2017-12-04T00:16:28Z

Build failed
Swift Test Linux Platform
Git Sha - 99710c535006b9a53c954df412a545a6cb906515

swift-ci · 2017-12-04T00:16:30Z

Build failed
Swift Test OS X Platform
Git Sha - 99710c535006b9a53c954df412a545a6cb906515

phausler · 2017-12-04T00:38:07Z

@swift-ci please test

swift-ci · 2017-12-04T00:38:30Z

Build failed
Swift Test OS X Platform
Git Sha - 47f1cf21b5eeaca32e73d843861122222570151b

swift-ci · 2017-12-04T00:39:27Z

Build failed
Swift Test Linux Platform
Git Sha - 47f1cf21b5eeaca32e73d843861122222570151b

zbowling · 2017-12-04T01:48:11Z

not sure how to invoke @swift-ci so it compiles this PR with foundation using this PR together swiftlang/swift-corelibs-foundation#1354

gparker42 · 2017-12-04T01:56:18Z

please test with the following pull request:
swiftlang/swift-corelibs-foundation#1354

@swift-ci please test

swift-ci · 2017-12-04T01:56:44Z

Build failed
Swift Test OS X Platform
Git Sha - 42a003164f08bdf5750b435c16fbc870532a4cb4

swift-ci · 2017-12-04T01:57:48Z

Build failed
Swift Test Linux Platform
Git Sha - 42a003164f08bdf5750b435c16fbc870532a4cb4

zbowling · 2017-12-04T02:56:32Z

doh. dependency in unit tests was broken. fixed

gparker42 · 2017-12-04T02:56:46Z

please test with the following pull request:
swiftlang/swift-corelibs-foundation#1354

@swift-ci please test

swift-ci · 2017-12-04T02:57:09Z

Build failed
Swift Test OS X Platform
Git Sha - 1fd55bb8dbd5f1d57b4dfeb13cf4106a86010a15

swift-ci · 2017-12-04T02:57:09Z

Build failed
Swift Test Linux Platform
Git Sha - 1fd55bb8dbd5f1d57b4dfeb13cf4106a86010a15

zbowling · 2017-12-04T03:14:03Z

Funny linker error in foundation:

../buildbot_linux/foundation-linux-x86_64/Foundation/Foundation/NSIndexSet.swift.o: In function `_T010Foundation10NSIndexSetC21_enumerateWithOptionsSiSgAA013NSEnumerationF0V_AA8_NSRangeV5rangexm9paramTypeq_m06returnK0q_x_SpyAA8ObjCBoolVGtc5blocktr0_lFySicq_x_AOtccfU_ySicfU_AI_ytTg5': ../buildbot_linux/foundation-linux-x86_64/Foundation/Foundation/NSIndexSet.swift.o:(.text+0x2b2f): undefined reference to `_T0s17_AnyCollectionBoxC6_indexs01_a5IndexC0_psAD_p_s5Int64V8offsetBytF'

Curious why this worked on a previous build.

Demangle says it's: Swift._AnyCollectionBox._index(Swift._AnyIndexBox, offsetBy: Swift.Int64) -> Swift._AnyIndexBox

I'm wondering if this is supposed to get generated by libswiftSwiftOnoneSupport.so and it's not picking it up now.

zbowling · 2017-12-04T03:20:18Z

updated this branch to current master. I'm wondering if this is just the branch foundation is on is out of sync with this one. should be ok to test again.

gparker42 · 2017-12-04T03:22:22Z

please test with the following pull request:
swiftlang/swift-corelibs-foundation#1354

@swift-ci please test

swift-ci · 2017-12-04T03:22:45Z

Build failed
Swift Test Linux Platform
Git Sha - d048dd6f5a429fe46651f648392f797c6da8a970

compnerd · 2017-12-05T17:48:48Z

lib/Driver/ToolChains.cpp

@@ -1016,6 +1016,17 @@ static void getRuntimeLibraryPath(SmallVectorImpl<char> &runtimeLibPath,
                          getPlatformNameForTriple(TC.getTriple()));
 }

+/// Get the runtime library link path with the target triple's specific arch
+/// library folder.
+static void getRuntimeLibraryPathWithArch(SmallVectorImpl<char> &runtimeLibPath,


Can you call this getRuntimeSharedLibraryPathWithArch to match getRuntimeStaticLibraryPathWithArch?

the non "WithArch" variant above this one calls doesn't use the "Shared" prefix. I could rename everything but it seemed more invasive.

compnerd · 2017-12-05T17:50:20Z

lib/Frontend/CompilerInvocation.cpp

+    // directory for searching for binaries. This is the same as the import
+    // path for modules.
+    SearchPathOpts.RuntimeLibraryPath = LibPath.str();
+  }
 }



Why not sink the LibPath calculation into the Triple.isOSDarwin() or else case respectively?

Sure. I'll look at writing this be more clear with one branch instead of two. I was over optimizing trying to avoid creating two string allocs for the non-Darwin path at the sake of some clarity.

Basically:

if Darwin: RuntimeLibraryPath = /lib/swift/(platform) else: RuntimeLibraryPath = /lib/swift/(platform)/(arch) RuntimeLibraryImportPath = /lib/swift/(platform)/(arch)

compnerd · 2017-12-05T17:52:34Z

Please test with the following pull request:
swiftlang/swift-corelibs-foundation#1354

@swift-ci please test

swift-ci · 2017-12-05T17:53:54Z

Build failed
Swift Test Linux Platform
Git Sha - db0242ead12a38e1852578724db7b01f8fff2450

swift-ci · 2017-12-05T17:57:44Z

Build failed
Swift Test OS X Platform
Git Sha - db0242ead12a38e1852578724db7b01f8fff2450

zbowling · 2017-12-05T19:18:18Z

I think one of the tests failed because it's not cleaning. A previous version created a symlink in an unfortunate spot that will confuse the "RUN: ln -s" statements in the test.

The other test I'm not sure why it fails. It's not failing that way locally on my Ubuntu 14.04 machine.

zbowling · 2017-12-05T19:27:58Z

validation-test/execution/interpret-with-dependencies-linux.swift

 // RUN: ln -s %t/libabc.so %t/rsrc/%target-sdk-name/%target-cpu
-// RUN: ln -s %platform-module-dir/../* %t/rsrc/%target-sdk-name/%target-cpu
+// RUN: ln -s %platform-module-dir/../%target_cpu/* %t/rsrc/%target-sdk-name/%target-cpu


doh. %target-cpu and not %target_cpu

zbowling · 2017-12-05T19:36:25Z

Seems like a regression to something discussed in #282

On platforms without fat binaries (anything but Darwin), we now build and install stdlibs for each arch in it's own folder. Namely lib/swift/<platform>/<arch> instead of lib/swift/<platform>.

zbowling · 2017-12-05T22:28:08Z

squashed and rebased this branch so I can update my fuchsia branch based on these commits

compnerd · 2017-12-12T06:00:54Z

Please test with the following pull request:
swiftlang/swift-corelibs-foundation#1354

@swift-ci please test

swift-ci · 2017-12-12T06:24:00Z

Build failed
Swift Test Linux Platform
Git Sha - 056a367

zbowling · 2017-12-12T10:16:50Z

Please test with the following pull request:
swiftlang/swift-corelibs-foundation#1354

@swift-ci please smoke test linux

zbowling · 2017-12-13T19:44:28Z

I'm not sure what this test is doing that is failing on the linux bot. The module maps generated are working on debain machines I've tested just fine but the test fails.

furkanozbay · 2017-12-18T17:37:52Z

@swift-ci please test

CodaFi · 2017-12-18T17:42:50Z

Drive-by log restoration. Don’t mind me

CodaFi · 2017-12-18T17:43:06Z

Please test with the following pull request:
swiftlang/swift-corelibs-foundation#1354

@swift-ci please smoke test Linux platform

compnerd · 2017-12-19T00:26:09Z

Do we want to split this up and get in part of the patch earlier? The build portion would be extremely useful for me as it allows me to build just part of the windows targets (e.g. just the x86_64 or ARM target).

zbowling · 2018-01-08T19:17:18Z

@compnerd it's very intertwined and already split up a lot so far. going to try and investigate this last test that is failing. it's some kind of intermediate thing testing some kind of module metadata that I don't really understand or something but the final product seems to working fine on all archs so it's weird.

alblue · 2018-01-26T13:55:56Z

@swift-ci please test linux

swift-ci · 2018-01-26T14:25:19Z

Build failed
Swift Test Linux Platform
Git Sha - 056a367

zbowling · 2018-02-07T02:58:40Z

I have some changes to swiftpm, libdispatch, and xctest (in additional to the current foundation change already) that go along with this change to teach them about the linux arch folder. Still not sure what is broken this unit test though.

CodaFi · 2019-11-21T20:44:41Z

Closing this due to age and inactivity, both here and on the corresponding Foundation PR.

zbowling commented Nov 28, 2017

View reviewed changes

gottesmm requested a review from Rostepher November 28, 2017 17:23

troughton mentioned this pull request Nov 29, 2017

[CMake] Windows stdlib cross-compilation fixes #13140

Merged

jrose-apple reviewed Dec 1, 2017

View reviewed changes

gparker42 reviewed Dec 1, 2017

View reviewed changes

zbowling mentioned this pull request Dec 4, 2017

Search arch specific lib directory swiftlang/swift-corelibs-foundation#1354

Closed

compnerd reviewed Dec 5, 2017

View reviewed changes

zbowling commented Dec 5, 2017

View reviewed changes

zbowling added 2 commits December 5, 2017 14:13

Fix multi-arch cross-compiling

807798e

On platforms without fat binaries (anything but Darwin), we now build and install stdlibs for each arch in it's own folder. Namely lib/swift/<platform>/<arch> instead of lib/swift/<platform>.

remove concept of swiftImageInspectionStatic/Shared libraries

056a367

zbowling force-pushed the multi_arch branch from a87b063 to 056a367 Compare December 5, 2017 22:14

jrose-apple mentioned this pull request Jan 29, 2018

build: depend on the architecture variant on non-MachO #14243

Closed

jrose-apple mentioned this pull request Feb 7, 2018

[driver] Implement a Windows toolchain for cross-compiling #13236

Merged

lanza mentioned this pull request Mar 8, 2018

Fix multi-arch cross-compiling #15085

Closed

CodaFi closed this Nov 21, 2019

allevato deleted the multi_arch branch October 7, 2020 19:37

Fix multi-arch cross-compiling #13108

Fix multi-arch cross-compiling #13108

Uh oh!

Conversation

zbowling commented Nov 28, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gparker42 Dec 1, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jrose-apple left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

compnerd commented Dec 3, 2017

Uh oh!

zbowling commented Dec 3, 2017

Uh oh!

zbowling commented Dec 3, 2017

Uh oh!

phausler commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

phausler commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

phausler commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

zbowling commented Dec 4, 2017

Uh oh!

gparker42 commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

zbowling commented Dec 4, 2017

Uh oh!

gparker42 commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

zbowling commented Dec 4, 2017

Uh oh!

zbowling commented Dec 4, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gparker42 commented Dec 4, 2017

Uh oh!

swift-ci commented Dec 4, 2017

Uh oh!

Choose a reason for hiding this comment

zbowling commented Nov 28, 2017 •

edited

Loading

gparker42 Dec 1, 2017 •

edited

Loading

zbowling commented Dec 4, 2017 •

edited

Loading