search

snowleopard / hadrian

Hadrian: a new build system for the Glasgow Haskell Compiler. Now merged into the GHC tree!
https://gitlab.haskell.org/ghc/ghc/tree/master/hadrian
MIT License
208 stars 39 forks source link

Allow caching of configure results. #607

Open AndreasPK opened 6 years ago

AndreasPK commented 6 years ago

Running configure takes a lot of time on windows.

Having a flag to use cached configure data for rebuilds would speed this up nicely.

I can run configure for GHC with -C myself. But adding the flag for the configure process of base is a different story. Ideally libraries would also use the same configure cache as ghc, so it would be faster even on the first build:

Here are the times for base and ghc itself on my machine. (SSD, skylake i7)

name ./configure ./configure -C (rerun) difference
ghc 61 s 18s -43
base 1m 40s 31s -1m 9s
snowleopard commented 6 years ago

@AndreasPK Good idea! Now that we got rid of ghc-cabal we should have a more fine-grain control over how packages are configured.

Let me tag @angerman @alpmestan @izgzhen -- I think they are more knowledgeable in how this corner of Hadrian could be improved.

alpmestan commented 6 years ago

@AndreasPK and I discussed this a little more on IRC. autoconf's caching mechanism handles a lot of the work. Andreas pointed out that we probably just need to run the "toplevel" configure script with -C and then pass --configure-option=-C to all the cabal packages when we're configuring them. This is expected to make a noticeable difference on Windows, but not that much on Linux.

angerman commented 6 years ago

Configure on macOS feels quite a bit slower than on Linux as well. For crosscompilation the slowness is just comical.

angerman commented 6 years ago

This used to be discussed on #ghc a few years ago, and I’m not sure about the exact details anymore, but I believe there was some fear around stale state. Maybe @bgamari remembers...

AndreasPK commented 6 years ago

I think it would be great to have the option. If it's a good default is indeed a different topic.

This used to be discussed on #ghc a few years ago, and I’m not sure about the exact details anymore, but I believe there was some fear around stale state. Maybe @bgamari remembers..

Personally I don't think this is a major issue if we delete the cache on make clean. There are already lot's of ways to break a working build tree with lingering build artefacts where make clean is required to fix it.

But I'm also perfectly happy if we make this available just as an optional setting.

angerman commented 6 years ago

I'm absolutely in favor of having this! I just wanted to add some history.