On Mon, Jan 29 2018, Ian Jackson wrote:

> Mattia Rizzolo writes ("Bug#844125: dgit: Built-in support for
>pbuilder [and 1 more messages]"):
>> [lots of explanation]
> Thanks, that is very helpful information.

Indeed.  Many thanks.

>> Anyway, isn't this discussion kind of moot?  I understood that for
>> reasons you're thinking of specifying `--debbuiltopts -b`, so you'd
>> get a binary only build and that one would discard the .orig as well
>> (as with -b the source don't appear on the .changes at all).
> Yes, well, technically it is moot for supporting pbuilder as a dgit
> build option.  But I think gbp users will probably want
> --build--products-dir more often than eg sbuild users, so proper
> pbuilder support means fixing #863582 (that --b-p-d does not always
> work properly) and #857316 (that it is not configurable).

Not fixing #863582 will make `dgit pbuilder` very confusing; on the
other hand, not fixing #857316 just makes it inconvenient.  So I think
this bug is blocked by only the former (as indeed the BTS metadata says

> It might be necessary for dgit to override some of the user's pbuilder
> config.  (And, as I said in my other mail, it might also be desirable
> for dgit to interrogate the pbuilder config so that it can honour it.)

Like Mattia, I'm very queasy about this, because pbuilder's config is
not declarative.  Indeed, everything seems to be shell scripts.

I haven't yet read your older mail where you came to the conclusion that
it would be necessary for dgit to parse pbuilder's config, but we should
be really sure it's necessary before going for that.

On Mon, Jan 29 2018, Mattia Rizzolo wrote:

> Then, there are other gotchas, that make the whole "ask pbuilder what
> its configuration is" kind of pointless: pbuilderrc is a shell script
> and therefore interpreted at runtime.  If a user wants to (I do for
> some things) it can compute values according to what I've been asked
> to do, so for example somebody may have options with different values
> according to what subcommand is being run.  But I believe this is very
> similar for sbuild as well, isn't it?

sbuild uses shell scripts to add things like tmpfs usage to the build,
but its configuration is a mixture of perl and ini, not shell.  So the
situation is slightly better.

Sean Whitton

Attachment: signature.asc
Description: PGP signature

Reply via email to