Compat fix for 6258 and fix deterministic flag #6295

dsyme · 2019-02-28T02:03:22Z

This makes desktop fsc.exe 32-bit again, see #6258. It includes #6223 which is not yet in master.

See #6258 (comment) for the rationale as to why.

The .NET Core fsc.exe remains 64-bit.

(Separately it would be good to know the perf implications either way, mainly out of curiosity as I don't think we'll change either choice. I guess if 64-bit is faster we could add a fscAnyCpu to the build and use it as the default desktop compiler going forward, leaving the other for stack-size/type-provider compat scenarios?)

… (64-bit) the default to launch in VS dotnet#6223

KevinRansom · 2019-02-28T02:20:47Z

@dsyme, longer term we will want to shift to the coreclr as the default compiler, and interactive, and keep the desktop around for legacy projects.

The coreclr is mostly going to be 64 bit, the 32 bit compiler is often not used or even installed.

dsyme · 2019-02-28T03:19:37Z

@dsyme, longer term we will want to shift to the coreclr as the default compiler, and interactive, and keep the desktop around for legacy projects. The coreclr is mostly going to be 64 bit, the 32 bit compiler is often not used or even installed.

Yes understood

KevinRansom · 2019-02-28T06:59:31Z

@dsyme

CompilerOptions\fsc\determinism (14) -- failed
CompilerOptions\fsc\determinism (15) -- failed
CompilerOptions\fsc\determinism (16) -- failed

dsyme · 2019-02-28T13:34:34Z

+++ CompilerOptions\fsc\determinism (14) +++
PRECMD: [fsc --deterministic                  dummy.fs && fsiAnyCPU copyArtifacts.fsx]
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
--------------------------------------------------------
Error from hosted compiler
Exit code: 0
Error:     0
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
--------------------------------------------------------
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
Peverify: [D:\a\1\s\tests\fsharpqa\Source\..\..\..\artifacts\bin\PEVerify\Release\net46\PEVerify.exe dummy.exe ]
Running: [ dummy.exe ]

POSTCMD: [fsiAnyCPU --nologo --quiet --exec binaryCompare.fsx true]
Expected D:\a\1\s\tests\fsharpqa\Source\CompilerOptions\fsc\determinism\dummy.exe and D:\a\1\s\tests\fsharpqa\Source\CompilerOptions\fsc\determinism\dummy2.exe to be same
Fail to execute the POSTCMD. FAIL
+++ CompilerOptions\fsc\determinism (15) +++
PRECMD: [fsc --deterministic --debug:portable dummy.fs && fsiAnyCPU copyArtifacts.fsx]
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
--------------------------------------------------------
Error from hosted compiler
Exit code: 0
Error:     0
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
--------------------------------------------------------
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
Peverify: [D:\a\1\s\tests\fsharpqa\Source\..\..\..\artifacts\bin\PEVerify\Release\net46\PEVerify.exe dummy.exe ]
Running: [ dummy.exe ]

POSTCMD: [fsiAnyCPU --nologo --quiet --exec binaryCompare.fsx true]
Expected D:\a\1\s\tests\fsharpqa\Source\CompilerOptions\fsc\determinism\dummy.exe and D:\a\1\s\tests\fsharpqa\Source\CompilerOptions\fsc\determinism\dummy2.exe to be same
Fail to execute the POSTCMD. FAIL
+++ CompilerOptions\fsc\determinism (16) +++
PRECMD: [fsc --deterministic --debug:embedded dummy.fs && fsiAnyCPU copyArtifacts.fsx]
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
--------------------------------------------------------
Error from hosted compiler
Exit code: 0
Error:     0
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
--------------------------------------------------------
Microsoft (R) F# Compiler version 10.2.3 for F# 4.5
Copyright (c) Microsoft Corporation. All Rights Reserved.
Peverify: [D:\a\1\s\tests\fsharpqa\Source\..\..\..\artifacts\bin\PEVerify\Release\net46\PEVerify.exe dummy.exe ]
Running: [ dummy.exe ]

POSTCMD: [fsiAnyCPU --nologo --quiet --exec binaryCompare.fsx true]
Expected D:\a\1\s\tests\fsharpqa\Source\CompilerOptions\fsc\determinism\dummy.exe and D:\a\1\s\tests\fsharpqa\Source\CompilerOptions\fsc\determinism\dummy2.exe to be same
Fail to execute the POSTCMD. FAIL

dsyme · 2019-02-28T14:02:48Z

It's weird, I can't repro this failure in the determinism tests locally

.\build net40 debug

cd tests\fsharpqa\Source\CompilerOptions\fsc\determinism

..\..\..\..\..\..\artifacts\bin\fsc\Debug\net46\fsc.exe --deterministic dummy.fs
fsiAnyCpu copyArtifacts.fsx
..\..\..\..\..\..\artifacts\bin\fsc\Debug\net46\fsc.exe --deterministic dummy.fs
fsiAnyCpu --quiet --exec binaryCompare.fsx true

This shows identical binaries are being produced

dsyme · 2019-02-28T14:03:19Z

@brettfo @KevinRansom Have you ever seen these tests fail?

KevinRansom · 2019-02-28T20:08:59Z

@dsyme … no.

dsyme · 2019-02-28T20:49:53Z

I think it's that

the HostedCompilerServer used by the fsharpqa tests to produce dummy.exe was running x64
the fsc.exe used by the same tests to produce dumym2.exe was running x86

Surprisingly the output binaries seem to differ slightly when the compiler runs as x86 v. x64.

@KevinRansom @brettfo any idea why this may be the case?

A vague guess is that this is because the PDB generation library writes slightly different entries

KevinRansom · 2019-02-28T20:51:52Z

If it's creating desktop pdbs then who knows. If it's portable pdbs, then they should be consistent.

dsyme · 2019-02-28T22:27:47Z

OK, well, the tests pass now, but there must be some small difference. I'll see if I can figure it out

…o x86

dsyme · 2019-03-01T00:00:22Z

@KevinRansom There was a bug in --deterministic - it was using hash on strings but these hashes are not consistent between x64 and x86 (indeed .NET hashes on strings are not even the same from day to day AFAIK)

I will add the fix here

KevinRansom · 2019-03-01T00:01:45Z

@dsyme , good eyes.

dsyme · 2019-03-01T00:29:48Z

@dsyme , good eyes.

2 hours looking at hexdumps of binaries....

cartermp · 2019-03-01T15:47:23Z

@KevinRansom @Pilchie We should consider servicing this in an RC service. I'll queue up a bug.

dsyme · 2019-03-01T16:20:50Z

BTW we could make the desktop fsc.exe and fsi.exe "anycpu, 32-bit preferred". They are currently x86.

I don't really see any technical downside to doing that - it means that if they happen to be launched as 64-bit processes then they will launch and run correctly.

* Make .NET Desktop fsi.exe 32-bit again and make Desktop fsiAnyCpu.exe (64-bit) the default to launch in VS #6223 * make fsc.exe 32-bit for compat * make fsc.exe 32-bit for compat * improve diagnostics * fix tests * fix determinism

Pilchie · 2019-03-04T21:27:54Z

I don't see a bug yet.

cartermp · 2019-03-04T21:57:08Z

I'll write one up - one-handed 🤕

dsyme added 3 commits February 28, 2019 01:58

Make .NET Desktop fsi.exe 32-bit again and make Desktop fsiAnyCpu.exe…

05d9a06

… (64-bit) the default to launch in VS dotnet#6223

make fsc.exe 32-bit for compat

0abe361

make fsc.exe 32-bit for compat

863c50c

KevinRansom approved these changes Feb 28, 2019

View reviewed changes

improve diagnostics

da9f58c

fix tests

8a14d5e

Merge branch 'master' of http://github.com/Microsoft/visualfsharp int…

e00e1db

…o x86

fix determinism

f41723a

cartermp changed the title ~~Compat fix for 6258~~ Compat fix for 6258 and fix deterministic flag Mar 1, 2019

dsyme merged commit 09a9870 into dotnet:master Mar 1, 2019

Compat fix for 6258 and fix deterministic flag #6295

Compat fix for 6258 and fix deterministic flag #6295

Uh oh!

Conversation

dsyme commented Feb 28, 2019

Uh oh!

KevinRansom commented Feb 28, 2019

Uh oh!

dsyme commented Feb 28, 2019

Uh oh!

KevinRansom commented Feb 28, 2019

Uh oh!

dsyme commented Feb 28, 2019

Uh oh!

dsyme commented Feb 28, 2019

Uh oh!

dsyme commented Feb 28, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

KevinRansom commented Feb 28, 2019

Uh oh!

dsyme commented Feb 28, 2019

Uh oh!

KevinRansom commented Feb 28, 2019

Uh oh!

dsyme commented Feb 28, 2019

Uh oh!

dsyme commented Mar 1, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

KevinRansom commented Mar 1, 2019

Uh oh!

dsyme commented Mar 1, 2019

Uh oh!

cartermp commented Mar 1, 2019

Uh oh!

dsyme commented Mar 1, 2019

Uh oh!

Pilchie commented Mar 4, 2019

Uh oh!

cartermp commented Mar 4, 2019

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

dsyme commented Feb 28, 2019 •

edited

Loading

dsyme commented Mar 1, 2019 •

edited

Loading