[WIP] Enhance -assume-single-threaded option #7421

mtake · 2017-02-13T12:30:26Z

(This PR was replaced with #7557)

This PR is part of SR-3945 which extends the scope of -assume-single-threaded option to value types.

This PR is not fully integrated with swift build system yet. The same environment variable SWIFT_ASSUME_SINGLE_THREADED is used at compile time to configure swift compiler to generate single threaded code and, as well as at runtime to configure swift native runtime for single threaded code.

To use this PR, set an arbitrary text to SWIFT_ASSUME_SINGLE_THREADED while running build-script as well as running your program. Without the variable defined, this PR should have no effect.

This change updates the branch mappings for the Swift-LLDB repository to match the LLVM & Clang model which LLDB is adopting.

The correct/canonical ConcreteType and ConcreteTypeSource are both stored in the Representative, but previously only the getter for the former was tranversing the chain.

This is the "canonical" representation of the type-level requirements of a protocol, and we intend to use it pervasively in the compiler.

The requirement signature is far more compact than looking at all the members.

When using the requirement signature in the archetype builder, not all nested types are resolved (i.e. added to their parent PA's NestedTypes map) when a protocol conformance is added. Equating a PA to a concrete type would previously rely on the NestedTypes map being complete, to recursively equate those nested types to the type implied by their parent being concrete. The lazier resolution now also includes resolving nested PA's to concrete types, if their parents are concrete.

This change depends on: swiftlang/swift-cmark#4

…ynthesized` again. This reverts part of #4038 which made the compiler consider it to be an `Explicit` conformance, breaking source code that was accepted in Swift 3.0 which declared a raw type as well as explicit conformance to `RawRepresentable` (reported as rdar://problem/30386658). While I'm here, a couple of spot fixes: - Ensure an enum's raw value exprs are type-checked before checking conformances of any of its extensions, since the RawRepresentable conformance derivation will blow up if the raw value exprs haven't been checked. Fixes an order dependency issue if `extension Foo: RawRepresentable {}` gets checked before `enum Foo: Int { ... }`. - Don't display the custom `enum_declares_rawrep_with_raw_type` diagnostic if the source location for the enum's inheritance clause is invalid, so that we don't emit a dislocated diagnostic.

Previously, these were only diagnosed in some situations, but the compiler is becoming more flexible about when things get fed into different places, and so can crop up elsewhere.

Default this option to its previous value (whether libxml2 was found), but allow us to independently turn it off.

This just moves a bunch of queries that used information on SILFunction to determine this property to just call a helper on SILFunctionType itself. Centralized logic is good. rdar://29791263

This is just refactoring code from DiagnoseUnreachable into a centralized location. rdar://29791263

… <rdar://problem/30271691>

…cheme

The requirement signature covers this functionality, when it matters.

…an a char pointer + length This avoids doing a strlen in lldb when calling this function. rdar://problem/30062026

…mparisons.

Matching two dependent member types has nothing to do with matching the bases; they're independent type parameters.

…nts. Rather than using "isEqual" to match same-type constraints among concrete types, use a full type matched to recursively decompose the structure. This allows us to support same-type constraints that end up being of the form X<T> == X<U>, where T and U are type parameters of some sort. Fixes rdar://problem/29333056.

…ariable (#7421)

…ironment variable (#7421)

…t runtime for single-threaded code (#7421)

swiftix · 2017-02-15T15:04:44Z

@swift-ci please smoke test

swiftix · 2017-02-15T15:05:05Z

@swift-ci please smoke test

swiftix · 2017-02-15T15:21:42Z

@mtake

The implementation with the environment variable seems interesting. We still need to check if it affects the performance of atomic cases too much due to the additional conditional branch instruction. If this is the case, we would need to go for the approach where we produce different binaries for the atomic and non-atomic runtime libraries.
BTW, we cannot easily test your PR on the CI bots, because it would require setting an environment variable, which is not currently possible using @swift-ci commands. But I'll try to test it locally on my machine till the end of the week. In general, we really need to have an option in the swift/util/build-script to enable the builds of non-atomic components (e.g. stdlib and runtime).
Could you locally rebase your PR to contain only your commits and then push force it using git push --force? Currently it contains about 35 commits from other people and those commits have nothing to do with your PR. Alternatively, you can submit a new PR, which would contain only your commits.

gottesmm · 2017-02-16T01:06:04Z

If you want to test on the PR bots, you can add in a commit on your branch a change to the preset that the bot will invoke or add an additional preset.

There is support in swift-ci for invoking arbitrary presets.

gottesmm · 2017-02-16T01:12:48Z

You can add a preset by looking in ./utils/build-presets.ini. I would just copy the section under:

[preset: buildbot_incremental,tools=RA,stdlib=RD,smoketest=macosx]

And create a preset maybe with the name:

[preset: buildbot_incremental,tools=RA,stdlib=RD,smoketest=macosx,single-thread]

and add your flag to that after the em-dash to ensure that it is passed to build-script-impl if you need to do so.

…ft into enhance-assume-single-threaded

…ariable (#7421)

…ironment variable (#7421)

…t runtime for single-threaded code (#7421)

…ft into enhance-assume-single-threaded

mtake · 2017-02-16T02:38:11Z

I merged upstream into my fork, created a new topic branch, applied all my changes in a single commit, pushed the branch to my fork, and tried to create a new PR from it. However, I discarded the PR since it still included many commits from other people. I need help from github expert for solving the problem.

swiftix · 2017-02-16T16:20:49Z

Here is what I usually do:

fork the original, i.e. apple/swift
clone my fork to my machine
Configure the upstream: https://help.github.com/articles/configuring-a-remote-for-a-fork/
create locally a branch for my feature
develop and commit on this branch
once I'm ready, I sync it with upstream like this (note that I do not merge!):
- git fetch upstream master, which receives the latest changes from apple/swift master
- git rebase upstream/master, which rebases your commits on top of those changes
push it to my branch on github using git push if it is a first time or eventually using git push --force if required
create a PR

Please let me know if it helps or if you need more help.

slavapestov · 2017-02-17T04:03:20Z

Instead of merging upstream into your branch, rebase your branch on top of upstream:

git fetch origin
git rebase -i origin/master
git push -f mygithubrepo HEAD:refs/heads/remote-branch-name

mtake · 2017-02-17T05:13:33Z

@swiftix Your suggestion to create a clean PR worked perfectly. Thanks!

Chris Bieneman and others added 30 commits February 10, 2017 11:42

Update the checkout script configuration for LLDB changes

e144637

This change updates the branch mappings for the Swift-LLDB repository to match the LLVM & Clang model which LLDB is adopting.

[AST] PotentialArchetype's ConcreteTypeSource is in the Representative.

f38f8ae

The correct/canonical ConcreteType and ConcreteTypeSource are both stored in the Representative, but previously only the getter for the former was tranversing the chain.

[AST] Store and serialize a protocol's requirement signature.

4491c84

This is the "canonical" representation of the type-level requirements of a protocol, and we intend to use it pervasively in the compiler.

[AST] Use the requirement signature in the ArchetypeBuilder.

857e4c5

The requirement signature is far more compact than looking at all the members.

Clean up handling of CMark library dependency.

618505e

This change depends on: swiftlang/swift-cmark#4

[AST] Diagnose recursive requirements whenever they appear.

80d1c94

Previously, these were only diagnosed in some situations, but the compiler is becoming more flexible about when things get fed into different places, and so can crop up elsewhere.

[Sema] Conformances only need to be verified as used once.

041d81f

[Parse] Fix crash in conditional compilation parsing (#7331)

8d9f736

Python lint update-checkout

4eb87a3

Python lint scale-test

72251fc

Python lint line-directive

22009b2

Python lint coverage-query-db

a7ac299

Python lint coverage-generate-data

77e587b

Python lint build-script

88160d2

42 * 42 is 1764, not 1746.

b73c85d

[CMake] Make SWIFT_HAVE_LIBXML an option.

a58548d

Default this option to its previous value (whether libxml2 was found), but allow us to independently turn it off.

Support for Open Existentials that do no take an address

aeac04d

[sil] Define SILFunctionType::isNoReturnFunction().

5524bae

This just moves a bunch of queries that used information on SILFunction to determine this property to just call a helper on SILFunctionType itself. Centralized logic is good. rdar://29791263

[sil] Add a new API SILModule::isNoReturnBuiltinOrIntrinsic(Identifier).

c24261f

This is just refactoring code from DiagnoseUnreachable into a centralized location. rdar://29791263

[Preset] Disable LLDB tests from PR testing and Swift package preset.…

b443352

… <rdar://problem/30271691>

[Update checkout] Use master branch for sub projects in master-next s…

05011d2

…cheme

[AST] Delete unnecessary generic environment use in ArchetypeBuilder.

21bbcc5

The requirement signature covers this functionality, when it matters.

fix docs build failure for open_existential_opaque

d9253d3

Mangling: isSwiftSymbol should get a null terminated string rather th…

1ca036d

…an a char pointer + length This avoids doing a strlen in lldb when calling this function. rdar://problem/30062026

[Archetype builder] Refactor same-type requirement checking slightly.

df926bc

[Archetype builder] Use addSameTypeRequirement() for concrete type co…

a473fcd

…mparisons.

[AST] Don't recursively match DependentMemberTypes.

8cd280c

Matching two dependent member types has nothing to do with matching the bases; they're independent type parameters.

mtake added 5 commits February 14, 2017 18:22

Change free-standing functions to member functions

17e0d40

A single runtime that initializes the atomicity with an environment v…

9a7e2f2

…ariable (#7421)

Flip the logic in the previous commit (#7421)

30fd5a1

A single compiler which initializes the default atomicity with an env…

934ae85

…ironment variable (#7421)

Stop using COMMON_C_FLAGS to configure swift compiler as well as swif…

e8d8333

…t runtime for single-threaded code (#7421)

swiftix self-assigned this Feb 15, 2017

swiftix requested review from gottesmm and swiftix February 16, 2017 01:06

swiftix removed their assignment Feb 16, 2017

mtake added 9 commits February 16, 2017 10:53

Enhance -assume-single-threaded option (#7421)

e043819

Merge branch 'enhance-assume-single-threaded' of github.com:mtake/swi…

1c2db74

…ft into enhance-assume-single-threaded

Extend the scope of -assume-single-threaded option to value types

a3880c0

Change free-standing functions to member functions

2b86594

A single runtime that initializes the atomicity with an environment v…

a4d4a86

…ariable (#7421)

Flip the logic in the previous commit (#7421)

c5bee03

A single compiler which initializes the default atomicity with an env…

af1df67

…ironment variable (#7421)

Stop using COMMON_C_FLAGS to configure swift compiler as well as swif…

9ce74ba

…t runtime for single-threaded code (#7421)

Merge branch 'enhance-assume-single-threaded' of github.com:mtake/swi…

b008d29

…ft into enhance-assume-single-threaded

mtake mentioned this pull request Feb 17, 2017

Enhance -assume-single-threaded option (SR-3945) #7557

Merged

mtake closed this Feb 17, 2017

mtake changed the title ~~[WIP] (Don't merge) Enhance -assume-single-threaded option~~ [WIP] Enhance -assume-single-threaded option Feb 17, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[WIP] Enhance -assume-single-threaded option #7421

[WIP] Enhance -assume-single-threaded option #7421

Uh oh!

mtake commented Feb 13, 2017 •

edited

Loading

Uh oh!

swiftix commented Feb 15, 2017

Uh oh!

swiftix commented Feb 15, 2017

Uh oh!

swiftix commented Feb 15, 2017

Uh oh!

gottesmm commented Feb 16, 2017

Uh oh!

gottesmm commented Feb 16, 2017

Uh oh!

mtake commented Feb 16, 2017 •

edited

Loading

Uh oh!

swiftix commented Feb 16, 2017

Uh oh!

slavapestov commented Feb 17, 2017

Uh oh!

mtake commented Feb 17, 2017 •

edited

Loading

Uh oh!

Uh oh!

[WIP] Enhance -assume-single-threaded option #7421

[WIP] Enhance -assume-single-threaded option #7421

Uh oh!

Conversation

mtake commented Feb 13, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

swiftix commented Feb 15, 2017

Uh oh!

swiftix commented Feb 15, 2017

Uh oh!

swiftix commented Feb 15, 2017

Uh oh!

gottesmm commented Feb 16, 2017

Uh oh!

gottesmm commented Feb 16, 2017

Uh oh!

mtake commented Feb 16, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

swiftix commented Feb 16, 2017

Uh oh!

slavapestov commented Feb 17, 2017

Uh oh!

mtake commented Feb 17, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

mtake commented Feb 13, 2017 •

edited

Loading

mtake commented Feb 16, 2017 •

edited

Loading

mtake commented Feb 17, 2017 •

edited

Loading