Use `Base.o_*` instead of raw `{#const O_*}` by GulinSS · Pull Request #339 · haskell/unix

GulinSS · 2025-04-25T16:38:26Z

stage1 cross compilers can use different values instead of system-defined. GHC JS Backend overrides these constants to be compatible with Node.js environment.

Details of investigation.

Interface stability tests have the following overridden constants:

  o_APPEND :: GHC.Internal.Foreign.C.Types.CInt
  o_BINARY :: GHC.Internal.Foreign.C.Types.CInt
  o_CREAT :: GHC.Internal.Foreign.C.Types.CInt
  o_EXCL :: GHC.Internal.Foreign.C.Types.CInt
  o_NOCTTY :: GHC.Internal.Foreign.C.Types.CInt
  o_NONBLOCK :: GHC.Internal.Foreign.C.Types.CInt
  o_RDONLY :: GHC.Internal.Foreign.C.Types.CInt
  o_RDWR :: GHC.Internal.Foreign.C.Types.CInt
  o_TRUNC :: GHC.Internal.Foreign.C.Types.CInt
  o_WRONLY :: GHC.Internal.Foreign.C.Types.CInt

Blocks: https://gitlab.haskell.org/ghc/ghc/-/merge_requests/13856

GulinSS · 2025-04-25T16:41:50Z

Set to draft while https://gitlab.haskell.org/ghc/ghc/-/merge_requests/13856 is in progress

`stage1` cross compilers could use different values instead of system-defined. GHC JS Backend change these constants to be compatible with Node.js environment.

GulinSS · 2025-04-25T18:17:01Z

It became actual after

https://gitlab.haskell.org/ghc/ghc/-/issues/25190
https://gitlab.haskell.org/ghc/ghc/-/issues/23697

Which were raised after

haskell/cabal#10366

These changes are for restoring consistency across cross compile targets with packages which have custom Setup.hs build rules.

GulinSS · 2025-04-27T05:09:21Z

All green https://gitlab.haskell.org/ghc/ghc/-/pipelines/110169.

More green at https://github.com/GulinSS/haskell-libraries-unix/actions/runs/14669542159

Bodigrim · 2025-04-28T22:57:30Z

+    -- them overridden.
+    -- For example GHC JS Backend provides its own constants
+    -- which should be used at the target of cross compilation
+    -- into Node.JS environment.


Could you reference a GHC source file which contain overriding logic for JS backend? It would help future readers.

Yes, we have a test which can be referenced for such overridings:

grep -rn testsuite -e 'o_CREAT' testsuite/tests/interface-stability/base-exports.stdout-javascript-unknown-ghcjs:13590: o_CREAT :: GHC.Internal.Foreign.C.Types.CInt testsuite/tests/interface-stability/base-exports.stdout:10544: o_CREAT :: GHC.Internal.Foreign.C.Types.CInt testsuite/tests/interface-stability/base-exports.stdout-mingw32:10812: o_CREAT :: GHC.Internal.Foreign.C.Types.CInt testsuite/tests/interface-stability/base-exports.stdout-ws-32:10544: o_CREAT :: GHC.Internal.Foreign.C.Types.CInt

So, files testsuite/tests/interface-stability/base-exports.stdout* can be useful to determine which targets have foreigns (exports? 🤔 ).

Actual overrides are collected at ./libraries/ghc-internal/src/GHC/Internal/System/Posix/Internals.hs. For example o_CREAT:

JS Backend

foreign import javascript unsafe "(() => { return h$base_o_creat; })" o_CREAT :: CInt

Other

foreign import ccall unsafe "HsBase.h __hscore_o_creat" o_CREAT :: CInt

Probably need create a [Note] there at sources?

Bodigrim · 2025-04-28T23:05:42Z

-fdOption2Int NonBlockingRead   = (#const O_NONBLOCK)
+fdOption2Int AppendOnWrite     = (Base.o_APPEND)
+fdOption2Int NonBlockingRead   = (Base.o_NONBLOCK)
 fdOption2Int SynchronousWrites = (#const O_SYNC)


Is there a reason for the inconsistency, beyond GHC exporting incomplete set? Could you please raise an issue on GHC side to provide a complete set in future? (We can leave unix source code inconsistent for the time being, but I'd like this to be rectified in future)

Bodigrim

I'm fine with accepting a tactical fix, but it raises questions about viability of hsc2hs for cross-compilation. What's the plan? CC @bgamari @mpickering

GulinSS · 2025-04-29T16:40:39Z

I'm fine with accepting a tactical fix, but it raises questions about viability of hsc2hs for cross-compilation. What's the plan? CC @bgamari @mpickering

@bgamari sent a response here. If I understand it correctly, tactical fix is OK but for further changes we need to do something with base if CLC OK. The main point is to prevent having a dependency on ghc-internal from unix and use base instead.

Bodigrim

LGTM, but we really need someone to resurrect CI before merging. I won't have time for this until next week unfortunately.

See actions/runner-images#11101

GulinSS · 2025-04-30T04:34:07Z

resurrect CI before merging

I did obvious fixes for the CI, but need some clues about NetBSD / OpenBSD.

hasufell · 2025-04-30T04:44:33Z

NetBSD and OpenBSD are constantly broken. I don't mind if we ignore them.

GulinSS · 2025-04-30T06:48:30Z

NetBSD is surprisingly fine. But OpenBSD is failed because 7.6 is a minimal version. We use image of 7.5 which was not updated by its author. Cirrus itself does not support OpenBSD OOTB.

I have no rights to merge, so, I am needed to wait.

GulinSS marked this pull request as draft April 25, 2025 16:38

Use Base.o_* instead of raw {#const O_*}

1b798b4

`stage1` cross compilers could use different values instead of system-defined. GHC JS Backend change these constants to be compatible with Node.js environment.

GulinSS marked this pull request as ready for review April 27, 2025 05:09

hasufell requested a review from Bodigrim April 28, 2025 13:22

hasufell approved these changes Apr 28, 2025

View reviewed changes

Bodigrim reviewed Apr 28, 2025

View reviewed changes

Bodigrim approved these changes Apr 29, 2025

View reviewed changes

GulinSS added 2 commits April 30, 2025 07:20

Update ci-js.yml

ce9f04e

See actions/runner-images#11101

Update ci-wasm32-wasi.yml

feca3f1

See actions/runner-images#11101

hasufell merged commit 47d5fc4 into haskell:master Apr 30, 2025

GulinSS deleted the wip/T24603 branch April 30, 2025 07:09

Uh oh!

Conversation

GulinSS commented Apr 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

GulinSS commented Apr 25, 2025

Uh oh!

GulinSS commented Apr 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

GulinSS commented Apr 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Bodigrim Apr 28, 2025

Choose a reason for hiding this comment

Uh oh!

GulinSS Apr 29, 2025 • 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!

GulinSS Apr 29, 2025

Choose a reason for hiding this comment

Uh oh!

GulinSS Apr 29, 2025

Choose a reason for hiding this comment

Uh oh!

Bodigrim Apr 28, 2025 • 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!

GulinSS Apr 29, 2025

Choose a reason for hiding this comment

Uh oh!

Bodigrim left a comment

Choose a reason for hiding this comment

Uh oh!

GulinSS commented Apr 29, 2025

Uh oh!

Bodigrim left a comment

Choose a reason for hiding this comment

Uh oh!

GulinSS commented Apr 30, 2025

Uh oh!

hasufell commented Apr 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

GulinSS commented Apr 30, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

GulinSS commented Apr 25, 2025 •

edited

Loading

GulinSS commented Apr 25, 2025 •

edited

Loading

GulinSS commented Apr 27, 2025 •

edited

Loading

GulinSS Apr 29, 2025 •

edited

Loading

Bodigrim Apr 28, 2025 •

edited

Loading

hasufell commented Apr 30, 2025 •

edited

Loading