If my reading of the diffs is correct (which I'm not sure it is), then it seems some identifiers in the hyperlinked source are missing type information. Looking at one of the changes Diff for file "src/Classes.html": 1194c1194,1198 < >

>

...

a -> b -> a forall a b. a -> b -> a

1281c1285,1289 < >

>

...

a -> b -> a forall a b. a -> b -> a

It seems like at certain identifiers, the types a -> b -> a and forall a b. a -> b -> a is now not being emitted in a place that they previously were. The html files are munged quite a bit to normalise spurious differences, but unfortunately this means that they don't render in a browser so well. If you have two builds of haddock (one prior to your patch, possibly even the one from your boot ghc bindist), you can run haddock-clean --html --hyperlinked-source utils/haddock/hypsrc-test/src/Classes.hs haddock-your-tree --html --hyperlinked-source utils/haddock/hypsrc-test/src/Classes.hs and then inspect/compare the rendered hyperlinked source file it produces (for the above command it would be in `src/Classes.html`) with your web browser. In the case of `Classes.hs`, it seems like perhaps some annotations on some of the `($)` operators in the following snippet are missing, and you can find out exactly which one by hovering over each of the `$` symbols and comparing with the output from a clean/unpatched haddock build. instance Plugh Either where plugh (Left a) _ = Right $ const a plugh (Right a) _ = Right $ const a plugh _ (Left b) = Left $ const b plugh _ (Right b) = Left $ const b These annotations are produced by walking the typechecked AST and asking for the types of each `Id` or `HsVar` etc. nodes, so anything your patched changed regarding how the AST is constructed could be to blame. On 24/05/31 10:41, Simon Peyton Jones wrote:

THanks. But the trouble is that I don't know whether to accept the change or not. Is it OK? How could I tell?

The point of a test is to fail when one has introduced a bug. This fails -- have I introduced a bug?

Thanks

S

On Fri, 31 May 2024 at 08:58, Zubin Duggal wrote:

...

You run this locally exactly the same as any other test, using

hadrian/build test --only=haddockHypsrcTest

or

TEST="haddockHypsrcTest" hadrian/build test

I recommend accepting the changes by passing the `-a` option to hadrian, and then looking over the changes locally using `git diff`.

The diffs reported by the test case itself can be confusing and I think (not sure) that they are in reversed order to what you would normally expect.

On 24/05/31 08:48, Simon Peyton Jones wrote:

...

Dear Haddock experts

I'm getting a failure on !12492 https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12492 that I can't understand or debug. Can someone guide me?

Failure is this: /builds/ghc/ghc/tmp/ghctest-2jo7oz8x/test spaces/testsuite/tests/haddock/haddock_testsuite/haddockHypsrcTest.run haddockHypsrcTest [bad exit code (2)] (normal)

A snip of the log is at the end of this message.

I want to

- understand what it does - repro it locally.

Trying the latter, I tried with my HEAD build and did bash$ cd testsuite/tests/haddock/haddock_testsuite bash$ make hypsrcTest

but that failed on almost all the tests it does, even though HEAD is presumably OK.

I'm stumped. Any ideas? The patch only changes the type checker!

Thanks

SImion

This is the log from !12492

Wrong exit code for haddockHypsrcTest()(expected 0 , actual 2 ) Stdout ( haddockHypsrcTest ): [1 of 7] Compiling Test.Haddock.Process (

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Process.hs, Test/Haddock/Process.o ) [2 of 7] Compiling Test.Haddock.Utils (

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Utils.hs, Test/Haddock/Utils.o ) [3 of 7] Compiling Test.Haddock.Config (

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Config.hs, Test/Haddock/Config.o ) [4 of 7] Compiling Test.Haddock (

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock.hs, Test/Haddock.o ) [5 of 7] Compiling Test.Haddock.Xhtml (

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Xhtml.hs, Test/Haddock/Xhtml.o ) [6 of 7] Compiling Main ( /builds/ghc/ghc/testsuite/../utils/haddock/hypsrc-test/Main.hs, Main.o ) [7 of 7] Linking hypsrc-test Haddock version 2.30.0, (c) Simon Marlow 2006 Ported to use the GHC API by David Waern 2006-2008 9.11.20240530 Generating documentation... Testing output files... Checking "src/TemplateHaskellSplices.html"... PASS Checking "src/Bug1091.html"... PASS Checking "src/Polymorphism.html"... PASS Checking "src/UsingQuasiquotes.html"... PASS Checking "src/Records.html"... PASS Checking "src/Quasiquoter.html"... PASS Checking "src/CPP.html"... PASS Checking "src/Operators.html"... PASS Checking "src/Literals.html"... PASS Checking "src/LinkingIdentifiers.html"... PASS Checking "src/Classes.html"... FAIL Checking "src/PositionPragmas.html"... PASS Checking "src/Types.html"... PASS Checking "src/TemplateHaskellQuasiquotes.html"... PASS Checking "src/Constructors.html"... FAIL Checking "src/Identifiers.html"... FAIL Diffing failed cases... Diff for file "src/Classes.html": 1194c1194,1198 < >

...

...

---

...

...

---

...

...

---

...

...

---

...

...

---

...

...

---

...

...

---

...

...

---

...

...

---

...

...

---

...

...

...

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

Thank you Zubin. In fact the issue has disappeared as mysteriously as it appeared. CI is green. I'm baffled, but disinclined to dig further! Simon On Fri, 31 May 2024 at 11:17, Zubin Duggal wrote:

I think its the annotations on `const` that changed, not on `$`, sorry.

On 24/05/31 15:38, Zubin Duggal wrote:

...

If my reading of the diffs is correct (which I'm not sure it is), then it seems some identifiers in the hyperlinked source are missing type information.

Looking at one of the changes

Diff for file "src/Classes.html": 1194c1194,1198 < >

...

><span class="annottext" >a -> b -> a forall a b. a -> b -> a </span >

1281c1285,1289 < >

...

><span class="annottext" >a -> b -> a forall a b. a -> b -> a </span >

It seems like at certain identifiers, the types a -> b -> a and forall a b. a -> b -> a is now not being emitted in a place that they previously were.

The html files are munged quite a bit to normalise spurious differences, but unfortunately this means that they don't render in a browser so well.

If you have two builds of haddock (one prior to your patch, possibly even the one from your boot ghc bindist), you can run

haddock-clean --html --hyperlinked-source utils/haddock/hypsrc-test/src/Classes.hs haddock-your-tree --html --hyperlinked-source utils/haddock/hypsrc-test/src/Classes.hs

and then inspect/compare the rendered hyperlinked source file it produces (for the above command it would be in `src/Classes.html`) with your web browser.

In the case of `Classes.hs`, it seems like perhaps some annotations on some of the `($)` operators in the following snippet are missing, and you can find out exactly which one by hovering over each of the `$` symbols and comparing with the output from a clean/unpatched haddock build.

instance Plugh Either where plugh (Left a) _ = Right $ const a plugh (Right a) _ = Right $ const a plugh _ (Left b) = Left $ const b plugh _ (Right b) = Left $ const b

These annotations are produced by walking the typechecked AST and asking for the types of each `Id` or `HsVar` etc. nodes, so anything your patched changed regarding how the AST is constructed could be to blame.

On 24/05/31 10:41, Simon Peyton Jones wrote:

...

THanks. But the trouble is that I don't know whether to accept the change or not. Is it OK? How could I tell?

The point of a test is to fail when one has introduced a bug. This fails -- have I introduced a bug?

Thanks

S

On Fri, 31 May 2024 at 08:58, Zubin Duggal wrote:

...

You run this locally exactly the same as any other test, using

hadrian/build test --only=haddockHypsrcTest

or

TEST="haddockHypsrcTest" hadrian/build test

I recommend accepting the changes by passing the `-a` option to hadrian, and then looking over the changes locally using `git diff`.

The diffs reported by the test case itself can be confusing and I think (not sure) that they are in reversed order to what you would normally expect.

On 24/05/31 08:48, Simon Peyton Jones wrote:

...

Dear Haddock experts

I'm getting a failure on !12492 https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12492 that I can't understand or debug. Can someone guide me?

Failure is this: /builds/ghc/ghc/tmp/ghctest-2jo7oz8x/test spaces/testsuite/tests/haddock/haddock_testsuite/haddockHypsrcTest.run haddockHypsrcTest [bad exit code (2)] (normal)

A snip of the log is at the end of this message.

I want to

- understand what it does - repro it locally.

Trying the latter, I tried with my HEAD build and did bash$ cd testsuite/tests/haddock/haddock_testsuite bash$ make hypsrcTest

but that failed on almost all the tests it does, even though HEAD is presumably OK.

I'm stumped. Any ideas? The patch only changes the type checker!

Thanks

SImion

This is the log from !12492

Wrong exit code for haddockHypsrcTest()(expected 0 , actual 2 ) Stdout ( haddockHypsrcTest ): [1 of 7] Compiling Test.Haddock.Process (

...

...

...

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Process.hs, Test/Haddock/Process.o ) [2 of 7] Compiling Test.Haddock.Utils (

...

...

...

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Utils.hs, Test/Haddock/Utils.o ) [3 of 7] Compiling Test.Haddock.Config (

...

...

...

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Config.hs, Test/Haddock/Config.o ) [4 of 7] Compiling Test.Haddock (

...

...

...

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock.hs, Test/Haddock.o ) [5 of 7] Compiling Test.Haddock.Xhtml (

...

...

...

...

/builds/ghc/ghc/testsuite/../utils/haddock/haddock-test/src/Test/Haddock/Xhtml.hs, Test/Haddock/Xhtml.o ) [6 of 7] Compiling Main ( /builds/ghc/ghc/testsuite/../utils/haddock/hypsrc-test/Main.hs, Main.o ) [7 of 7] Linking hypsrc-test Haddock version 2.30.0, (c) Simon Marlow 2006 Ported to use the GHC API by David Waern 2006-2008 9.11.20240530 Generating documentation... Testing output files... Checking "src/TemplateHaskellSplices.html"... PASS Checking "src/Bug1091.html"... PASS Checking "src/Polymorphism.html"... PASS Checking "src/UsingQuasiquotes.html"... PASS Checking "src/Records.html"... PASS Checking "src/Quasiquoter.html"... PASS Checking "src/CPP.html"... PASS Checking "src/Operators.html"... PASS Checking "src/Literals.html"... PASS Checking "src/LinkingIdentifiers.html"... PASS Checking "src/Classes.html"... FAIL Checking "src/PositionPragmas.html"... PASS Checking "src/Types.html"... PASS Checking "src/TemplateHaskellQuasiquotes.html"... PASS Checking "src/Constructors.html"... FAIL Checking "src/Identifiers.html"... FAIL Diffing failed cases... Diff for file "src/Classes.html": 1194c1194,1198 < >

...

><span class="annottext" >a -> b -> a forall a b. a -> b -> a </span >

---

...

><span class="annottext" >a -> b -> a forall a b. a -> b -> a </span >

---

...

><span class="annottext" >b -> a -> b forall a b. a -> b -> a </span >

---

...

><span class="annottext" >b -> a -> b forall a b. a -> b -> a </span >

---

...

><span class="annottext" >([Int] -> Int) -> (Norf -> [Int]) -> Norf -> Int forall b c a. (b -> c) -> (a -> b) -> a -> c </span >

---

...

><span class="annottext" >(String -> IO ()) -> (Int -> String) -> Int -> IO () forall b c a. (b -> c) -> (a -> b) -> a -> c </span >

---

...

><span class="annottext" >Int -> Int -> Int </span ><a href="Identifiers.html#foo" >

---

...

><span class="annottext" >(String -> IO ()) -> (Int -> String) -> Int -> IO () forall b c a. (b -> c) -> (a -> b) -> a -> c </span >

---

...

><span class="annottext" >Int -> Int </span ><a href="Identifiers.html#quux" >

---

...

><span class="annottext" >(String -> IO ()) -> (Int -> String) -> Int -> IO () forall b c a. (b -> c) -> (a -> b) -> a -> c </span >

---

...

><span class="annottext" >Int -> Int -> Int -> Int </span ><a href="Identifiers.html#norf" >

...

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs