[bug#34466] Fwd: Re: [bug#34466] [PATCH] gnu: Add coq-equations.

(sorry, sending this again because I forgot to CC the bugtracker)


Hi Julien.

Thank you for your thorough comments, I really appreciate it!
I hope that the quality of my patches increase in time so please bear with me for a bit :)

On 13-02-19 12:04, Julien Lepiller wrote:
> Le 2019-02-13 11:40, Dan Frumin a écrit :
>> ---
>>  gnu/packages/coq.scm | 40 ++++++++++++++++++++++++++++++++++++++++
>>  1 file changed, 40 insertions(+)
>>
>> diff --git a/gnu/packages/coq.scm b/gnu/packages/coq.scm
>> index fa02f85cd..494c3404b 100644
>> --- a/gnu/packages/coq.scm
>> +++ b/gnu/packages/coq.scm
>> @@ -500,3 +500,43 @@ work on a decision procedure for the equational
>> theory of an extension of the
>>  sigma-calculus by Abadi et al.  The library is completely written in Coq and
>>  uses Ltac to synthesize the substitution operation.")
>>        (license license:bsd-3))))
> 
> Hi Dan, thanks for this patch! I have some comments below to help improve the
> quality of future patches, as well as two questions you need to answer before
> I can push that patch.
> 
>> +
>> +(define-public coq-equations
>> +  (package
>> +    (name "coq-equations")
>> +    (synopsis "Equations - a function definition plugin")
> 
> Could you put this just before the description field, so this package definition
> looks more like the rest of guix? A better synopsis would be "Function definiton
> plugin for coq" I think.

I've reordered the fields to match the rest of the package definitions in guix.

> 
>> +    (version "1.2-beta")
> 
> Why a beta version? We try to stick to stable releases in Guix. Please add a
> comment explaining the reason.
> 

Oh, I just wanted to use the latest released version. I can replace it with the version 1.1 while I wait for the non-beta 1.2 releas.

>> +    (source (origin
>> +              (method url-fetch)
>> +              (uri (string-append
>> + "https://github.com/mattam82/Coq-Equations/archive/v"
>> +                    version "-8.8.tar.gz"))
>> +              (file-name (string-append name "-v" version "8.8.tar.gz"))
> 
> We cannot use auto-generated tarballs from github, because we found that
> they sometimes get regenerated in an unreproducible way, so it breaks the
> checksum test. You can use this instead:
> 
> (method git-fetch)
> (uri (git-reference
>         (url "https://githu.com/mattam82/Coq-Equations.git")
>         (commit (string-append "v" version "-8.8"))))
> 
> and update the sha256 accordingly. As an added bonus, this means that we
> can always fetch from the software heritage in case the repo disappears
> one day :)

I was not aware of that. How can I get the sha256 hash in this case? Normally I would do `guix download <url>`.

> 
>> +              (sha256
>> +               (base32
>> "1j7yarhddk2c2l4b6h8g5n0xz5vfy1bqmgh832g01di5gjwshy3f"))))
>> +    (build-system gnu-build-system)
>> +    (native-inputs
>> +     `(("findlib" ,ocaml)
> 
> ocaml doesn't provide findlib directly, ocaml-findlib does. What do you
> want to do here?

Sorry, I think I was confused here. I've corrected this in the updated patch.

> 
>> +       ("coq"     ,coq)
>> +       ("camlp5"  ,camlp5)))
>> +    (arguments
>> +     `(#:test-target "test-suite"
>> +       #:phases
>> +       (modify-phases %standard-phases
>> +         (replace 'configure
>> +           (lambda* (#:key outputs #:allow-other-keys)
>> +             (invoke "coq_makefile" "-f" "_CoqProject" "-o" "Makefile")))
>> +         (replace 'install
>> +           (lambda* (#:key outputs #:allow-other-keys)
>> +             (setenv "COQLIB" (string-append (assoc-ref outputs
>> "out") "/lib/coq/"))
>> +             (invoke "make"
>> +                     (string-append "COQLIB=" (assoc-ref outputs "out")
>> +                                    "/lib/coq/")
>> +                     "install"))))))
> 
> Please make sure that these two phases both return #t.

I thought that `invoke' automatically checks that the return code is correct?

> 
>> +    (description "Equations provides a notation for writing programs
>> +by dependent pattern-matching and (well-founded) recursion in Coq. It
>> +compiles everything down to eliminators for inductive types, equality
>> +and accessibility, providing a definitional extension to the Coq
>> +kernel.")
> 
> Please make sure that each sentence is separated by two spaces.
> `guix lint coq-equations` should be able to tell you about it.
> 
>> +    (home-page "https://mattam82.github.io/Coq-Equations/")
>> +    (license license:lgpl2.1)))
> 
> Thanks again! I really only need an answer for the beta version and
> ocaml/findlib questions. I can take care of the rest, but I would
> appreciate it if you could do it yourself ;)

Le 2019-02-13 12:36, Dan Frumin a écrit :
> (sorry, sending this again because I forgot to CC the bugtracker)
> 
> 
> Hi Julien.
> 
> Thank you for your thorough comments, I really appreciate it!
> I hope that the quality of my patches increase in time so please bear
> with me for a bit :)
> 
> On 13-02-19 12:04, Julien Lepiller wrote:
>> 
>> We cannot use auto-generated tarballs from github, because we found 
>> that
>> they sometimes get regenerated in an unreproducible way, so it breaks 
>> the
>> checksum test. You can use this instead:
>> 
>> (method git-fetch)
>> (uri (git-reference
>>         (url "https://githu.com/mattam82/Coq-Equations.git")
>>         (commit (string-append "v" version "-8.8"))))
>> 
>> and update the sha256 accordingly. As an added bonus, this means that 
>> we
>> can always fetch from the software heritage in case the repo 
>> disappears
>> one day :)
> 
> I was not aware of that. How can I get the sha256 hash in this case?
> Normally I would do `guix download <url>`.

you can always try to build the package, which will fail and tell you
what the correct hash was. I forgot, the filename should be:

(file-name (git-file-name name version))

>> 
>> Please make sure that these two phases both return #t.
> 
> I thought that `invoke' automatically checks that the return code is 
> correct?

It does, and when the code is not correct, it throws an exception. When 
it is
correct, it returns #<unspecified>, so we need to explicitly return #t.

Thank you! I'll try to build the package and take care of the rest 
before pushing
your patch later today.

Hi Julien,

> When [invoke's child return status] is correct, [invoke] returns #<unspecified>, so we need to explicitly return #t.

That's not true.  invoke returns #t in the successful case--I designed it that way mostly to get rid of the many vestigal #t's in phases.

(I hope that in a future version of guix, the necessity of using #t to end phases will be gone entirely)

Le 2019-02-13 13:39, Danny Milosavljevic a écrit :
> Hi Julien,
> 
>> When [invoke's child return status] is correct, [invoke] returns 
>> #<unspecified>, so we need to explicitly return #t.
> 
> That's not true.  invoke returns #t in the successful case--I designed
> it that way mostly to get rid of the many vestigal #t's in phases.
> 
> (I hope that in a future version of guix, the necessity of using #t to
> end phases will be gone entirely)

Ah, thanks for the info! So I guess we don't need to end any phase with 
#t
in that package definition, correct?

On Wed, 13 Feb 2019 13:45:43 +0100
Julien Lepiller <julien@lepiller.eu> wrote:

> Ah, thanks for the info! So I guess we don't need to end any phase with 
> #t
> in that package definition, correct?

Correct.