Re: lots of fatal build system bugs

[Ingo Schwarze]

Hi Ralph,

Ralph Corderoy wrote on Tue, Mar 22, 2022 at 10:32:20AM +0000:
> Ingo Schwarze wrote:

>> It would no doubt be better if i tested more often and did not let so
>> many issues accumulate.  :-/

> Once the current problems are fixed, would it be worth you automating
> the pull and build of new commits so the first breakage is flagged
> earlier?

Actually i hate automated systems where you can't commit when some
machine thinks it might not build or not adhere to some style guide.
That's somewhat fragile, always extremely massive bloat, and often
more of a nuisance than it helps.  Then again, i do realize you don't
argue at all for the insanity of preventing commits.

> Either a program like https://www.buildbot.net or something ad hoc.

Actually, i do have a ten-line shell script to run the build in the
git tree.  Maybe i should add ten more lines to that including the groff
"make dist" step and the steps needed to update and build the OpenBSD port
(make makesum, make patch, make configure, make, make fake, make package,
all will full logs being stored in appropriate, well-defined places).
That would probably help me to run it more often without spending too
much time on it.  Maybe not a bad idea at all.

> After a quick glance, you might just forward the errors on for
> others to sort out and it

Not sure that would be a fair thing to do - what would *you* do with a
report like this: "The OpenBSD 'make fake' stage fails with these 300
lines of messages [... long stretch of log messages ...].  None of these
messages provide any clue to me of what exactly fails, let alone why,
except that much is built there that shouldn't really be built at this
stage, while some artifacts that should be built have not been built
for unidentified reasons.  Maybe you can find out what is going on.
There is no apparent problem with running the command 'make' inside the
groff git tree."  That would have been my report for the issues 1 and 2
following your advice.  :-o

So it might be more productive that i investigate myself
if and when i find the time.

An immediate heads-up "something might be amiss here, i don't know
yet what exactly, wait a second before piling more commits on
top in this region" might make sense, even though immediate reverts
would probably be overkill and fixing it after the fact will usually
suffice.

> would avoid having to untangle problems caused by multiple commits.

Yes, that is indeed quite desirable.

> Regarding BSD's make, didn't some Linux distributions used to
> package it just as they do for byacc?

Quite possible, but apart from mandoc(1), i know very little about
which Linux distros provides packages of which BSD-style software.
I suspect there are people on this list who know more about various
packaging system on Linux distros than i do.  :)

Yours,
  Ingo