Leo Famulari <l...@famulari.name> writes:
> On Mon, Aug 01, 2016 at 07:34:49PM +0000, ng0 wrote:
>> * gnu/packages/irc.scm (ircii): New variable.
>
> Cool!
>
>> + (snippet
>> + '(begin
>> + (substitute* "bsdinstall"
>> + (("/bin/strip") "strip")
>> + (("/bin/cp") "cp")
>> + (("/bin/chmod") "chmod")
>> + (("/etc/chown") "chown")
>> + (("/bin/chgrp") "chgrp")
>> + (("/bin/mkdir") "mkdir")
>> + (("/bin/rm") "rm")
>> + (("/bin/mv") "mv"))))))
>
> Does anyone else think we should put this origin snippet in a build phase
> instead? Or, should we leave it in the origin, where it will be applied
> to the source code provided by `guix build --source ircii`?
I think it should be in snippet. The 'substitute*' doesn't introduce
any guix specified modifitions, it removes the hardcoded paths.
>
>> + (delete 'check))))
>
> Using '#:tests? #f' is preferred instead of deleting the check phase.
>
>> + (inputs
>> + `(("libiconv" ,libiconv)
>> + ("ncurses" ,ncurses)
>> + ("openssl" ,openssl)))
>
> I noticed that the built package does refer to libiconv:
>
> ---
> $ guix gc --references $(./pre-inst-env guix build ircii)
> /gnu/store/0kml8g9fix69v00afv59ngf4lgfr1565-openssl-1.0.2h
> /gnu/store/9maps38bsia0wcxm82h0v0p2dxyn8j35-ircii-20151120
> /gnu/store/9nifwk709wajpyfwa0jzaa3p6mf10vxs-gcc-4.9.3-lib
> /gnu/store/m9vxvhdj691bq1f85lpflvnhcvrdilih-glibc-2.23
> /gnu/store/xl19qrfzga52vrvp4ncccwjlnrjqwj95-ncurses-6.0
> /gnu/store/ykzwykkvr2c80rw4l1qh3mvfdkl7jibi-bash-4.3.42
> ---
>
> But, libiconv is used:
>
> ---
> $ strings /gnu/store/9maps38bsia0wcxm82h0v0p2dxyn8j35-ircii-20151120/bin/irc
> | grep iconv
> iconv_close
> iconv
> iconv_open
> iconv_open@@GLIBC_2.2.5
> iconv@@GLIBC_2.2.5
> iconv_close@@GLIBC_2.2.5
> ---
>
> Perhaps libiconv should be propagated?
The iconv functions are provided by glibc, libiconv is for systems that
do not use glibc. libiconv should be removed from inputs.
