On CI

Fri Feb 19 12:20:07 UTC 2021

Recompilation avoidance

I think in order to cache more in CI, we first have to invest some time in
fixing recompilation avoidance in our bootstrapped build system.

I just tested on a hadrian perf ticky build: Adding one line of *comment*
in the compiler causes

   - a (pretty slow, yet negligible) rebuild of the stage1 compiler
   - 2 minutes of RTS rebuilding (Why do we have to rebuild the RTS? It
   doesn't depend in any way on the change I made)
   - apparent full rebuild the libraries
   - apparent full rebuild of the stage2 compiler

That took 17 minutes, a full build takes ~45minutes. So there definitely is
some caching going on, but not nearly as much as there could be.
I know there have been great and boring efforts on compiler determinism in
the past, but either it's not good enough or our build system needs fixing.
I think a good first step to assert would be to make sure that the hash of
the stage1 compiler executable doesn't change if I only change a comment.
I'm aware there probably is stuff going on, like embedding configure dates
in interface files and executables, that would need to go, but if possible
this would be a huge improvement.

On the other hand, we can simply tack on a [skip ci] to the commit message,
as I did for https://gitlab.haskell.org/ghc/ghc/-/merge_requests/4975.
Variants like [skip tests] or [frontend] could help to identify which tests
to run by default.

Lean

I had a chat with a colleague about how they do CI for Lean. Apparently, CI
turnaround time including tests is generally 25 minutes (~15 minutes for
the build) for a complete pipeline, testing 6 different OSes and
configurations in parallel:
https://github.com/leanprover/lean4/actions/workflows/ci.yml
They utilise ccache to cache the clang-based C++-backend, so that they only
have to re-run the front- and middle-end. In effect, they take advantage of
the fact that the "function" clang, in contrast to the "function" stage1
compiler, stays the same.
It's hard to achieve that for GHC, where a complete compiler pipeline comes
as one big, fused "function": An external tool can never be certain that a
change to Parser.y could not affect the CodeGen phase.

Inspired by Lean, the following is a bit inconcrete and imaginary, but
maybe we could make it so that compiler phases "sign" parts of the
interface file with the binary hash of the respective subcomponents of the
phase?
E.g., if all the object files that influence CodeGen (that will later be
linked into the stage1 compiler) result in a hash of 0xdeadbeef before and
after the change to Parser.y, we know we can stop recompiling Data.List
with the stage1 compiler when we see that the IR passed to CodeGen didn't
change, because the last compile did CodeGen with a stage1 compiler with
the same hash 0xdeadbeef. The 0xdeadbeef hash is a proxy for saying "the
function CodeGen stayed the same", so we can reuse its cached outputs.
Of course, that is utopic without a tool that does the "taint analysis" of
which modules in GHC influence CodeGen. Probably just including all the
transitive dependencies of GHC.CmmToAsm suffices, but probably that's too
crude already. For another example, a change to GHC.Utils.Unique would
probably entail a full rebuild of the compiler because it basically affects
all compiler phases.
There are probably parallels with recompilation avoidance in a language
with staged meta-programming.

Am Fr., 19. Feb. 2021 um 11:42 Uhr schrieb Josef Svenningsson via ghc-devs <
ghc-devs at haskell.org>:

> Doing "optimistic caching" like you suggest sounds very promising. A way
> to regain more robustness would be as follows.
> If the build fails while building the libraries or the stage2 compiler,
> this might be a false negative due to the optimistic caching. Therefore,
> evict the "optimistic caches" and restart building the libraries. That way
> we can validate that the build failure was a true build failure and not
> just due to the aggressive caching scheme.
>
> Just my 2p
>
> Josef
>
> ------------------------------
> *From:* ghc-devs <ghc-devs-bounces at haskell.org> on behalf of Simon Peyton
> Jones via ghc-devs <ghc-devs at haskell.org>
> *Sent:* Friday, February 19, 2021 8:57 AM
> *To:* John Ericson <john.ericson at obsidian.systems>; ghc-devs <
> ghc-devs at haskell.org>
> *Subject:* RE: On CI
>
>
>    1. Building and testing happen together. When tests failure
>    spuriously, we also have to rebuild GHC in addition to re-running the
>    tests. That's pure waste.
>    https://gitlab.haskell.org/ghc/ghc/-/issues/13897
>    <https://nam06.safelinks.protection.outlook.com/?url=https://gitlab.haskell.org/ghc/ghc/-/issues/13897&data=04%7C01%7Csimonpj@microsoft.com%7C3d503922473f4cd0543f08d8d48522b2%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637493018301253098%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0=%7C3000&sdata=FG2fyYCXbacp69Q8Il6GE0aX+7ZLNkH1u84NA/VMjQc=&reserved=0>
>    tracks this more or less.
>
> I don’t get this.  We have to build GHC before we can test it, don’t we?
>
> 2 .  We don't cache between jobs.
>
> This is, I think, the big one.   We endlessly build the exact same
> binaries.
>
> There is a problem, though.  If we make **any** change in GHC, even a
> trivial refactoring, its binary will change slightly.  So now any caching
> build system will assume that anything built by that GHC must be rebuilt –
> we can’t use the cached version.  That includes all the libraries and the
> stage2 compiler.  So caching can save all the preliminaries (building the
> initial Cabal, and large chunk of stage1, since they are built with the
> same bootstrap compiler) but after that we are dead.
>
> I don’t know any robust way out of this.  That small change in the source
> code of GHC might be trivial refactoring, or it might introduce a critical
> mis-compilation which we really want to see in its build products.
>
> However, for smoke-testing MRs, on every architecture, we could perhaps
> cut corners.  (Leaving Marge to do full diligence.)  For example, we could
> declare that if we have the result of compiling library module X.hs with
> the stage1 GHC in the last full commit in master, then we can re-use that
> build product rather than compiling X.hs with the MR’s slightly modified
> stage1 GHC.  That **might** be wrong; but it’s usually right.
>
> Anyway, there are big wins to be had here.
>
> Simon
>
>
>
>
>
>
>
> *From:* ghc-devs <ghc-devs-bounces at haskell.org> *On Behalf Of *John
> Ericson
> *Sent:* 19 February 2021 03:19
> *To:* ghc-devs <ghc-devs at haskell.org>
> *Subject:* Re: On CI
>
>
>
> I am also wary of us to deferring checking whole platforms and what not. I
> think that's just kicking the can down the road, and will result in more
> variance and uncertainty. It might be alright for those authoring PRs, but
> it will make Ben's job keeping the system running even more grueling.
>
> Before getting into these complex trade-offs, I think we should focus on
> the cornerstone issue that CI isn't incremental.
>
>    1. Building and testing happen together. When tests failure
>    spuriously, we also have to rebuild GHC in addition to re-running the
>    tests. That's pure waste.
>    https://gitlab.haskell.org/ghc/ghc/-/issues/13897
>    <https://nam06.safelinks.protection.outlook.com/?url=https://gitlab.haskell.org/ghc/ghc/-/issues/13897&data=04%7C01%7Csimonpj@microsoft.com%7C3d503922473f4cd0543f08d8d48522b2%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637493018301253098%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0=%7C3000&sdata=FG2fyYCXbacp69Q8Il6GE0aX+7ZLNkH1u84NA/VMjQc=&reserved=0>
>    tracks this more or less.
>    2. We don't cache between jobs. Shake and Make do not enforce
>    dependency soundness, nor cache-correctness when the build plan itself
>    changes, and this had made this hard/impossible to do safely. Naively this
>    only helps with stage 1 and not stage 2, but if we have separate stage 1
>    and --freeze1 stage 2 builds, both can be incremental. Yes, this is also
>    lossy, but I only see it leading to false failures not false acceptances
>    (if we can also test the stage 1 one), so I consider it safe. MRs that only
>    work with a slow full build because ABI can so indicate.
>
> The second, main part is quite hard to tackle, but I strongly believe
> incrementality is what we need most, and what we should remain focused on.
>
> John
> _______________________________________________
> ghc-devs mailing list
> ghc-devs at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devs/attachments/20210219/90921cc9/attachment.html>