Bug#875684: Bug#870669: libidn: Make source package bootstrappable

[Manuel A. Fernandez Montecelo]

Control: tags -1 + pending


Hi,

2017-11-19 23:49 Manuel A. Fernandez Montecelo:
> 2017-11-04 23:09 Manuel A. Fernandez Montecelo:
> > 2017-09-13 14:19 Helmut Grohne:
> >
> > > Though I'd much rather see libidn go. Most rdeps but hesiod have moved
> > > on to libidn2-0.

Are you sure about this, Helmut?  There are lots of packages that
build-depend on libidn11-dev (like 55), very few on libidn2-dev and
libidn2-0-dev (there was a rename), like 10 in total.

So unless "build-rdeps --old" is not accurate, I think that most deps
have not migrated yet.

It looks like this package will continue with us for many months,
possibly years and at least one stable release, so the more important to
get this sorted out for a package which is part of the initial
debootstrap set.


> > Since it's unlikely to happen very soon (from a comment by upstream in
> > this report), I think that it shouldn't stop this being applied.
>
> I am preparing a NMU for this fix.
>
> To the maintainers: if you don't want it applied please speak soon, so I
> don't waste time on a fix that will be reverted :)

Since there has been no reply from the maintainers to neither of #870669
nor #875684 since reported (4.5 and 3+ months respectively), and almost
a month since my last e-mail announcing my intention to NMU, I think
that I am going to upload straight away rather than upload to delayed.

I wouldn't normally be in a hurry when uploading such a thing, but I
have more time to attend to any issue caused with the upload in the next
week than in the last days of the year.

.debdiff attached, but since the package is under collab-maint I am
going to push the changes there too.

I imported the changes of 1.33-2, which hadn't been added (or pushed) to
the repo in collab-maint.


Lastly, a note about my changes related to #875684 (using help2man when
cross-building).

I decided to disable it for cross-builds rather than building twice,
instead of the original one proposed, for the following reasons:

- Some builds in the last 2-3 years took 20 or 30 minutes in release
 architectures (armel, mips and mipsel), and in several ports arches
 took 30 mins to 1h, and in some extreme case 5h with m68k (built in
 qemu I think).  Normally one would cross-build with a fast machine,
 but if somebody does it with hardware not that powerful (e.g. a small
 arm board) or in emulated environments, it's a significant increase.
 (Even going from 5 to 10 minutes is significant, IMO).

- Unlike in other packages, the file is right there, and it's there for
 reasons like this, I imagine, otherwise it serves no purpose.

- Even if I know that it will make your (Helmut) checks with diffoscope
 complain about this file, at a more fundamental level, I don't think
 that going the extra mile to build a file which is a dump of --help in
 the command line makes any sense.

 This is done to satisfy the "requirement" in Debian to ship manpages,
 but I find it an odd practice (a static page saying "undocumentaed,
 try --help" would probably be better) and even harmful, because it
 creates problems for cross-building.

- And more importantly, the original patch for that bug made me scratch
 my head a bit and it complicates things, while I prefer to either make
 minimal or straightforward changes or, in the best cases, simplify the
 packaging, not making it more difficult to understand.

- If it's really really important to do it in that way, we can always
 change it later, it's not like this is a definitive change.


Cheers.
--
Manuel A. Fernandez Montecelo <address@hidden>

libidn_1.33-2.1.debdiff
Description: Text document