Re: [ANNOUNCE] GHC 8.0.1 source tarball available

Páli Gábor János writes:

2016-05-14 13:06 GMT+02:00 Karel Gardas :

...

On 05/14/16 11:28 AM, Ben Gamari wrote:

...

The pragmatist in me wants to answer 1) yes, 2) no, although I do dislike the idea of distributing binaries that weren't derived from the associated source tarball.

I guess all other Linuxes naturally use gnu make as `make' and Windows in msys too so only non-GNU/non-Linux systems should be affected and from those only FreeBSD has caught this.

Yes, that is possible. I do not know either Solaris or OpenBSD well enough, but I suspect they might have GNU make(1) installed in their paths as `make` or their default make(1) can understand GNU-style Makefiles. FreeBSD has BSD make(1), which is the default, and this cannot comprehend the GNU-style files at all.

Anyhow, in my humble opinion, it is a bad practice the hardwire the name of the make tool in the sources.

Indeed, this was my mistake. I'll try to be more careful of this in the future.

...

If this is true, then I would recommend "no" to both points and leave the fix in 8.0 branch for 8.0.2...

Well, in theory, FreeBSD is still a Tier-1 platform, so every release should just build fine without any further efforts. I am also aware of the fact I am considered a minority here, and that this is just a minor technical problem that could wait for some undetermined time. However, personally, I would be quite disappointed if this promise was broken.

Yes, you are right. FreeBSD is tier-1 and we have committed to ensure these work out-of-the-box. I had neglected to consider this in my previous assessment of the situation. In light of this I think we have little choice but to throw out these binaries and re-spin. Thankfully I have held off on pushing the tag until the last possible moment. I think at the moment we should include the following in the new release, * the haddock $(MAKE) fix * the patch vendorising the alabaster theme for haddock's documentation * the patch fixing the clean rule for haddock's documentation * the patch ensuring haddock documentation is built for ghc's `all` target * D2224, which splits ghc-boot to avoid unnecessary transitive dependencies in template-haskell (which otherwise would have necessitated a prompt 8.0.2 release) * A small fix for PPC which fixes crashes in threaded programs In the interest of risk minimization I think that is all we should merge.

I am sorry and apologize that I found this bug after the release was tagged, but I did not have the chance to test it before it was considered a final release.

No need to apologize; I'm glad you brought up the issue. The release will go out when it's ready. Cheers, - Ben