Don't repair install-plans of `alex` !

[andreasabel]

alex < 3.2.2 is listed as broken on matrix.hackage.

E.g. the build log for https://matrix.hackage.haskell.org/#/package/alex/3.2.1/ghc-8.2.2@1616116046 says:

Build profile: -w ghc-8.2.2 -O1
In order, the following will be built (use -v for more details):
 - alex-3.2.1 {alex-3.2.1-inplace} (exe:alex) (first run)
...
Preprocessing executable 'alex' for alex-3.2.1..
setup: The program 'happy' is required but it could not be found

Maybe the fix is to add build-tools: happy. However, this could have been omitted on purpose. Note that instead the tarball contains preprocessed versions of the lexer (.x) and parser (.y) definitions:

dist
└── build
    └── alex
        └── alex-tmp
            ├── Parser.hs
            └── Scan.hs
src
├── AbsSyn.hs
...
├── Parser.y
├── Scan.hs
├── Scan.x
...

So maybe alex-3.2.1 was relying on the preprocessed versions to be picked up by cabal?

P.S.: How auspicious, getting issue number #321 to repair alex-3.2.1. :)

[phadej]

They cannot be repaired, even you get happy available (by having it in PATH, which matrix doesn't so we catch missing build-tool-depends):

alex-3.2.1 master % cabal build -w ghc-8.0.2
Resolving dependencies...
Build profile: -w ghc-8.0.2 -O1
In order, the following will be built (use -v for more details):
 - alex-3.2.1 (exe:alex) (first run)
[1 of 1] Compiling Main             ( /codetmp/alex/alex-3.2.1/dist-newstyle/build/x86_64-linux/ghc-8.0.2/alex-3.2.1/setup/setup.hs, /codetmp/alex/alex-3.2.1/dist-newstyle/build/x86_64-linux/ghc-8.0.2/alex-3.2.1/setup/Main.o )
Linking /codetmp/alex/alex-3.2.1/dist-newstyle/build/x86_64-linux/ghc-8.0.2/alex-3.2.1/setup/setup ...
Configuring alex-3.2.1...
Preprocessing executable 'alex' for alex-3.2.1...
alex: /codetmp/alex/alex-3.2.1/././AlexTemplate-ghc: openFile: does not exist (No such file or directory)

The build-type: Custom Setup.hs is not working. Later alex are build-type: Simple.

FYI: build-tool-depends can (should) have version range.

[andreasabel]

Thanks for the insight, @phadej !

Based on the verdict The program 'happy' is required but it could not be found I did the following experiment on alex-3.2.1:

• I removed my happy
• cabal get alex-3.2.1
• cd alex-3.2.1
• cabal --storedir=$PWD install -w ghc-8.2.2 then failed with the same error (missing happy)
• I added build-tool: happy to alex-3.2.1 and retried install
• This succeeded, installing happy locally on the way

Following these results, I revised alex-3.2.1, only to see matrix.hackage failing with a different error:

setup: The program 'alex' is required but it could not be found

Well, I still had alex in my PATH, which poisoned my experiment.

So, I am going to undo my revision.

Also, in the light of https://github.com/haskell/cabal/issues/7808 I am no longer sure what the FAIL(pkg) verdict of matrix.hackage means. Because the following experiments might have different outcome.

1. cabal install alex-NNN (remote installation)
2. cabal get alex-NNN; cd alex-NNN; cabal install (local installation)
3. cabal get alex-NNN; cd alex-NNN; cabal build (local build)
4. In principle, installing alex-NNN as a build-tool dependency could have even have another semantics than the above three.

Perhaps matrix.hackage should make the setup of its experiment explicit, so its verdicts can be interpreted more precisely.

[phadej]

alex needing alex is exactly why https://github.com/simonmar/alex/pull/120 was done. Repairing build-type: Custom alexes are not worth your time.

[andreasabel]

I wrote:

So, I am going to undo my revision.

So I thought, but looks I went down a one-way street. My attempt to revert the revision is failing with:

Cannot remove existing legacy-tool dependency on 'happy' in executable 'alex' component

Well, extensionally my revision did not make much of a difference..., so this is not a catastrophy.

P.S.: How auspicious, getting issue number #321 to repair alex-3.2.1. :)

Yeah, not so auspicious... (More like famous last words...) :-D