Re: [PATCH] gnu: Add sendmail

[Alex Vong <alexvong1995@gmail.com>]

Alex Kost <alezost@gmail.com> writes:

> John Darrington (2016-09-17 12:11 +0200) wrote:
>
>> On Sat, Sep 17, 2016 at 05:38:26PM +0800, Alex Vong wrote:
> [...]
>>      >      > + "contrib/mmuegel" "devtools/bin/configure.sh")
>>      >      > +               (("/bin/sh") (which "bash")))
>>      >      > +
>>      >      > +             (substitute* "devtools/bin/Build"
>>      >      > + (("SHELL=/bin/sh") (string-append "SHELL=" (which "bash"))))
>>      >      > +             #t))
>>      >      I think the `#t' is not neccessary here, since `substitute*' uses
>>      >      `substitute', which will either return #t or throw an exception.
>>      >
>>      > WTF??  Didn't you complain earlier this week when I *didn't* put #t in
>>      > exactly this
>>      > scenario??
>>      >
>>      Yes, I am a different Alex :)
>>      Also, it seems we are not being consistent here, sometimes we put `#t'
>>      after `substitute*', sometimes we don't. Anyone has an idea?
>>
>> I did raise some suggestions in my earlier posts.  But again I don't
>> have any strong
>> opinion.
>
> I have a strong opinion: if a docstring of a procedure says what value
> it returns, we can rely on it, otherwise we should not guess what value
> will be returned.  In case of 'substitute*' (and 'substitute'), the
> returned value is not specified, so I think if a phase ends with
> 'substitute*', we should (or even must) add #t after it.

I see your point that one should not be relying on undocumented
features, which I agree. But I also see an alternative: to make
'substitute*' either return true or throw an exception and document
it. I think the heart of the problem is scheme is "untyped", so we rely
on the documentation. What do you think?