build failures for posix-api-0.1.0.0

[hvr]

Not sure if you noticed but there's build failures showing up at https://matrix.hackage.haskell.org/#/package/posix-api

(and this affects its reverse dependencies such as
https://matrix.hackage.haskell.org/#/package/ping)

[andrewthad]

Unfortunately, this package relies on a not-yet-released version of hsc2hs. In order to build it (or anything that depends on it), I tell cabal to use a local copy of hsc2hs. I should document this in the readme.

[hvr]

@andrewthad that means as soon as the respective hsc2hs version is released (or we know the version it'll have), we'll have to declare the lower version bound required in the .cabal file (via build-tools or build-tools-depends) via a metadata revision. Fwiw, in general Hackage is supposed to be self-contained, i.e. buildable with released dependencies (libraries as well as buildtools or toolchains) as otherwise it gets really complicated to ensure accurate metadata (which is esssential to a good user experience; our (simplified) mantra is that it's always preferable to fail during build-plan solving than to fail during compiling/typechecking).

[andrewthad]

Ah, I was not aware of build-tool-depends. Since this libraries only supports cabal 2.2 and up, I should be able to use build-tool-depends, which is good. I have a PR making this requirement explicit. I would go ahead and merge it, but it unfortunately prevents the library from building even when I use --with-hsc2hs to explicitly feed in a version of hsc2hs that agrees with the lower bound.

[andrewthad]

Also, with bd647c6, I've made the current shortcoming more clear in a readme.

[hvr]

Since this libraries only supports cabal 2.2 and up, I should be able to use build-tool-depends, which is good.

For the record, build-tools: hsc2hs >= x && < y is desugared into build-tool-depends: hsc2hs:hsc2hs >= x && < y; And there's been quite a few HAckage metadata revisions filling in missing build-tools: ... declarations for various core tools... as people tend to forget to specify them; we'll probably have to add a warning to cabal check to reduce this common mistake :-)

it unfortunately prevents the library from building even when I use --with-hsc2hs to explicitly feed in a version of hsc2hs that agrees with the lower bound.

yes, this is somewhat of a known issue because in v2-build we currently let cabal take care of procurring and building the required version and so far this hasn't been a problem because of that invariant I mentioned earlier that we generally don't upload things with dependencies on libs/tools that aren't released yet ;-)