Re: bug#30680: [racket-users] Using Racket's raco on on Guix(SD)

[Christopher Lemmer Webber]

Timothy Sample writes:

> Christopher Lemmer Webber <address@hidden> writes:
>
>> Konrad Hinsen writes:
>>
>>> In my tests, all packages ended up working, but performance is indeed
>>> worse than with a Racket installation outside of Guix.
>>>
>>> It would be nice if someone with more knowledge of Racket internals
>>> could give a hint or two for debugging this issue!
>>>
>>> Konrad.
>>
>> I'm posting a bug bounty on this issue: if someone can fix this I will
>> pay them $250 USD.  I don't have the time or knowledge enough of Racket
>> internals to do so myself.
>
> I have discovered a few things, but I’m not sure how to fix the
> underlying problem(s).
>
> The reason Racket is trying to recompile the OpenSSL files is because of
> a hash mismatch.  This can be seen by enabling debugging output:
>
>     $ PLTSTDERR=debug raco setup openssl
>
> Which says a lot of things, but most interestingly it says:
>
> --------------------------------
> ...
> compiler/cm: checking: 
> /gnu/store/jx0bkmaafb8fq0mqs5ywgnxq8rbpn8j1-racket-6.12/share/racket/collects/openssl/libcrypto.rkt
> compiler/cm: different src hash... (5d9ca57f3e267d956c7b5e62578467beb8ccc1d2 
> 4d21ac412723fbf33f97669c2f73f0e9367f4510)
> compiler/cm: maybe-compile-zo starting 
> /gnu/store/jx0bkmaafb8fq0mqs5ywgnxq8rbpn8j1-racket-6.12/share/racket/collects/openssl/libcrypto.rkt
> compiler/cm:   start-compile: 
> /gnu/store/jx0bkmaafb8fq0mqs5ywgnxq8rbpn8j1-racket-6.12/share/racket/collects/openssl/libcrypto.rkt
> compiler/cm:   compiling 
> /gnu/store/jx0bkmaafb8fq0mqs5ywgnxq8rbpn8j1-racket-6.12/share/racket/collects/openssl/libcrypto.rkt
> open-output-file: cannot open output file
>   path: 
> /gnu/store/jx0bkmaafb8fq0mqs5ywgnxq8rbpn8j1-racket-6.12/share/racket/collects/openssl/compiled/tmp15340167971534016797570
>   system error: Read-only file system; errno=30
>   context...:
> ...
> --------------------------------
>
> This hash mismatch is caused by grafting.  When the package is built,
> the path to OpenSSL gets hard-coded in a source file.  The SHA-1 hash
> for this file is stored in its “.dep” file.  When the output is
> grafted, the source file gets updated with a new OpenSSL path, but the
> hash does not get updated.  This makes Racket think that the cached
> bytecode file is incorrect (even though it was likely grafted too),
> and it tries to recompile it.  It fails because it tries to write this
> new bytecode file to the store.

Interesting... I hadn't even considered grafting.  (I still wonder why
it's even trying to open *any* file in the store for output though...)

> I double checked this by trying with an ungrafted Racket, and got better
> results.  (There was still a warning about writing to the store, but it
> seemed less significant.)

Cool!

> The only thing I can think of for a fix would be to patch Racket to be
> more lenient with bytecode files in the store.  That is, ignore hash
> mismatches in store-files.  I might give this a try later tonight if
> nobody has any better ideas.
>
> -- Tim

BTW, some examples of packages where I've had trouble, in case it helps
with testing:

 - Raart
 - Gregor
 - crypto (seemed to work last time, not sure why it wasn't working before)

Though at this point I also can't do just "raco setup" on a local
package either, but maybe resolving this issue will fix that.