fairywren is generating bogus BASE_BACKUP commands

Thomas Munro pointed out this failure to me on fairywren:

https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=fairywren&dt=2022-01-21%2020%3A10%3A22

He theorizes that I need some perl2host magic in there, which may well
be true. But I also noticed this:

# Running: pg_basebackup --no-sync -cfast --target
server:/home/pgrunner/bf/root/HEAD/pgsql.build/src/bin/pg_basebackup/tmp_check/tmp_test_Ag8r/backuponserver
-X none
pg_basebackup: error: could not initiate base backup: ERROR:
unrecognized target: "server;C"

"server" is a valid backup target, but "server;C" is not. And I think
this must be a bug on the client side, because the server logs the
generated query:

2022-01-21 20:53:11.618 UTC [8404:10] 010_pg_basebackup.pl LOG:
received replication command: BASE_BACKUP ( LABEL 'pg_basebackup base
backup', PROGRESS, CHECKPOINT 'fast', MANIFEST 'yes',
TABLESPACE_MAP, TARGET 'server;C', TARGET_DETAIL
'\\tools\\msys64\\home\\pgrunner\\bf\\root\\HEAD\\pgsql.build\\src\\bin\\pg_basebackup\\tmp_check\\tmp_test_Ag8r\\backuponserver')

So it's not that the server parses the query and introduces gibberish
-- the client has already introduced gibberish when constructing the
query. But the code to do that is pretty straightforward -- we just
call strchr to find the colon in the backup target, and then
pnstrdup() the part before the colon and use the latter part as-is. If
pnstrdup were failing to add a terminating \0 then this would be quite
plausible, but I think it shouldn't. Unless the operating sytem's
strnlen() is buggy? That seems like a stretch, so feel free to tell me
what obvious stupid thing I did here and am not seeing...

--
Robert Haas
EDB: http://www.enterprisedb.com