[Monotone-devel] Re: illegal instruction again [was: Re: monotone 0.11 test failures]

[graydon hoare]

Derek Scherger wrote:

> Test 24 seems to fail, but I haven't consciously enabled i18n or 
> anything so I'm not too worried about that.

yes. I recently upgraded my development system to FC2 and it appeared 
there too. some other people have seen it also. it appears that autotest 
developers decided NLS is a "nuisance", if you look in the testsuite 
file it expands to, you see they always clobber NLS support environment 
variables back to the "C" locale. charming, eh?

# NLS nuisances.
for as_var in \
  LANG LANGUAGE LC_ADDRESS LC_ALL LC_COLLATE LC_CTYPE LC_IDENTIFICATION \
  LC_MEASUREMENT LC_MESSAGES LC_MONETARY LC_NAME LC_NUMERIC LC_PAPER \
  LC_TELEPHONE LC_TIME
do
  if (set +x; test -z "`(eval $as_var=C; export $as_var) 2>&1`"); then
    eval $as_var=C; export $as_var
  else
    $as_unset $as_var
  fi
done

for the time being I will hack around this using the fact that idna 
accepts the "CHARSET" variable as well, which they're not perturbing. 
but this is a tenuous and unsatisfying hack.

> Test 26 just seems to hang up and not come back... on a clean 0.11 
> source tree. Interestingly test 23 seems to indicate that it can 
> communicate though?!?

yeah, there was a strange (i.e. I never understood it) issue with test 
23 not liking the way I put the netsync server into the background. I 
fiddled with it and got it working again in my development tree (it is 
now test 22, since I removed tests for the old networking code).

in general, I'm pretty disappointed with how hard it is to control the 
environment and prerequisites (running processes, especially) under 
autotest. it's really absurd. I thought autotest might be a marginal 
step up from dejagnu -- at least it's all one file -- but it is 
seemingly very weak too.

maybe qmtest will fare better. I'm shopping around.

> BTW, do I need to do something to see logging output? I'm attempting to 
> emit a message with the following, which I don't really understand very 
> well and nothing seems to get printed.

by default, L(...) logging goes to an internal circular buffer. if you 
want to view that while it's happening, run with --verbose.

conversely, the P(...) logging form goes to output unless you run with 
--quiet. P is mnemonic for "progress", indicators you think the user 
will want to see. L(...) is for things you suspect the user doesn't 
normally care about, unless there's a bug or misbehavior.

-graydon