Solaris "sed" versus pre-v13 plpython tests

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: pgsql-hackers(at)lists(dot)postgresql(dot)org
Cc: Noah Misch <noah(at)leadboat(dot)com>, hofstetter(at)jomasoft(dot)ch
Subject: Solaris "sed" versus pre-v13 plpython tests
Date: 2022-08-31 22:25:01
Message-ID: 165561.1661984701@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

I noticed that buildfarm member margay, which just recently started
running tests on REL_12_STABLE, is failing the plpython tests in that
branch [1], though it's happy in v13 and later. The failures appear
due to syntax errors in Python "except" statements, and it's visible
in some of the tests that we are failing to convert ancient Python
"except" syntax to what Python 3 wants:

diff -U3 /home/marcel/build-farm-14/buildroot/REL_12_STABLE/pgsql.build/src/pl/plpython/expected/python3/plpython_types_3.out /home/marcel/build-farm-14/buildroot/REL_12_STABLE/pgsql.build/src/pl/plpython/results/python3/plpython_types.out
--- /home/marcel/build-farm-14/buildroot/REL_12_STABLE/pgsql.build/src/pl/plpython/expected/python3/plpython_types_3.out 2022-08-31 22:29:51.070597370 +0200
+++ /home/marcel/build-farm-14/buildroot/REL_12_STABLE/pgsql.build/src/pl/plpython/results/python3/plpython_types.out 2022-08-31 22:29:53.053544840 +0200
@@ -400,15 +400,16 @@
import marshal
try:
return marshal.loads(x)
-except ValueError as e:
+except ValueError, e:
return 'FAILED: ' + str(e)
$$ LANGUAGE plpython3u;
+ERROR: could not compile PL/Python function "test_type_unmarshal"
+DETAIL: SyntaxError: invalid syntax (<string>, line 6)

So it would seem that regress-python3-mangle.mk's sed command to
perform this transformation isn't working on margay's sed.

We've had to hack regress-python3-mangle.mk for Solaris "sed" before
(cf commit c3556f6fa). This seems to be another instance of that
same crummy implementation of '*' patterns. I suppose Noah missed
this angle at the time because the problematic pattern had already
been removed in v13 and up (45223fd9c). But it's still there in v12.

I am not completely sure why buildfarm member wrasse isn't failing
similarly, but a likely theory is that Noah has got some other sed
in his search path there.

I confirmed on the gcc farm's Solaris 11 box that the pattern
doesn't work as expected with /usr/bin/sed:

tgl(at)gcc-solaris11:~$ echo except foo, bar: | sed -e 's/except \([[:alpha:]][[:alpha:].]*\), *\([[:alpha:]][[:alpha:]]*\):/except \1 as \2:/g'
except foo, bar:

We could perhaps do this instead:

$ echo except foo, bar: | sed -e '/^ *except.*,.*: *$/s/, / as /g'
except foo as bar:

It's a little bit more brittle, but Python doesn't allow any other
commands on the same line does it?

Another idea is to try to find some other sed to use. I see that
configure already does that on most platforms, because ax_pthread.m4
has "AC_REQUIRE([AC_PROG_SED])". But if there's still someone
out there using --disable-thread-safety, they might think this is
an annoying movement of the build-requirements goalposts.

regards, tom lane

[1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=margay&dt=2022-08-31%2020%3A00%3A05

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Zhihong Yu 2022-08-31 23:08:52 Re: cataloguing NOT NULL constraints
Previous Message Alvaro Herrera 2022-08-31 22:19:10 Re: cataloguing NOT NULL constraints