Ian, Guillem,
On 19 September 2019 at 14:51, Ian Jackson wrote:
| Dirk Eddelbuettel writes ("Possible doc package side-effect from going
source-only upload"):
| > Maybe someone on the list can help with a sharp insight before I go trying.
| >
| > The r-base source package (for the R system and language) has a somewhat
| > cobbled together debian/rules [1], mostly of my making over the last 20+
| > years since I helped Doug more and more and eventually took it over. I
| > apologize for the rough shape it is in, but hey, it works. Mostly. Read on.
| ....
| > So presumably the dependency graph within debian/rules is wrong. Would
| > anybody here know
| > - either a failsafe idiom forcing the right thing to happen
| > - or a more efficient way
| > to ensure the binary-arch is built before binary-all? Should I force it? Is
| > that wasteful? Is there a recommended way?
|
| You could take the first part of the binary-arch target and split it
| out into something that both binary-arch and binary-indep depend on.
| That would probably "fix" this problem.
Yes, my thinking was along those lines, but so far less refined. Will try this.
| Holger Levsen writes ("Re: Possible doc package side-effect from going
source-only upload"):
| > not really that helpful of a comment, but I think the rules file would
| > be a lot more readable if you'd dropped all the old commented out code
| > in it.
|
| I agree with this.
I did not see Holger's post as he did not CC me (and I had a crazy day, have
not check d-devel archives). I agree in the abstract and feel the same from
time ot time -- but in practice the comments are 'lab notes' for myself and
have been rather helpful a few times. And as I maintain the file they'll
stay for as long as they help me.
| > (and then I think^wbelieve your arch-all problem could be solved by
| > switching to dh style...)
|
| This would be a good idea. It is quite some effort but I think you
| would be rewarded with significantly lower maintenance burden.
Agreed in principle. Finding time to do it the right is the issue. I am
maintainer for well over 100 packages most of which are cookie-cutter CRAN
packages and alike -- they all use dh. A handful of older / larger packages
are still older-school.
On 19 September 2019 at 16:03, Guillem Jover wrote:
| On Thu, 2019-09-19 at 07:15:43 -0500, Dirk Eddelbuettel wrote:
| > So presumably the dependency graph within debian/rules is wrong. Would
| > anybody here know
| >
| > - either a failsafe idiom forcing the right thing to happen
| > - or a more efficient way
| >
| > to ensure the binary-arch is built before binary-all? Should I force it? Is
| > that wasteful? Is there a recommended way?
|
| I've just skimmed over the rules, but I indeed see several problematic
| constructs there:
|
| - «build: build-arch» that should include build-indep too.
Ok.
| - install-arch has build-arch as a dependency, but install-indep
| has make-indep, this seems inconsitent and possibly wrong.
I will take a look.
| - Some of the prerequisites are not protected behind stamp files, so
| they will get executed multiple times, which seems counter to using
| stamp files (although I find stamp files to be somewhat of an
| anti-pattern :). Stuff like:
|
| barrier: prereq-1 barrier
| barrier-stamp:
|
| but this is in a way covered already by the next point.
I used them in the 1990s when most if not all debian/rules files had, I never
really had they feeling they worked all that well. Use of them seems to have
disappeared over time.
| - It seems generally parallel unsafe, as many targets declare what
| should be serially executed as parallely-executed, as in:
|
| target: prereq-a prereq-b
|
| can execute both prereqs at the same time. The way to serialize
| them is:
|
| prereq-1: prereq-2
| target: prereq-1
Fair. Do we build packages with 'make -j ...' now?
| Not sure whether fixing all the above will fix your problem, but this
| is something that should be done non the less IMO. :)
All good hints. Much appreciated :)
Dirk
--
http://dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org
