changes in pgport etc doesn't cause client programs to be relinked

From: Andres Freund <andres(at)anarazel(dot)de>
To: pgsql-hackers(at)postgresql(dot)org
Subject: changes in pgport etc doesn't cause client programs to be relinked
Date: 2021-10-26 18:04:54
Message-ID: 20211026180454.xcjmu3kwmn3tka57@alap3.anarazel.de
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi,

our make dependencies currently are insufficient to trigger client binaries to
be relinked when pgport/pgcommon/libpq/... changes.

To reproduce, you can use something like:
touch ~/src/postgresql/src/port/snprintf.c && make -j48 2>&1|grep 'gcc.*pgbench'
which won't show anything, whereas e.g.
touch ~/src/postgresql/src/include/postgres_fe.h && make -j48 2>&1|grep 'gcc.*pgbench'
will.

The reason for that is that currently client programs only have order-only
dependencies on pgport:

pgbench: $(OBJS) | submake-libpq submake-libpgport submake-libpgfeutils
$(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $(at)$(X)

which will dutifully cause pgport to be rebuilt. But then we won't actually
rebuild $client_program, because it's just an order-only dependency [1].

The same problem does *not* exist for the backend, because there we add
pgport/pgcommon to OBJS, which will cause them to be proper dependencies.

This does explain some mysterious issues I had over the years with changes
occasionally requiring a clean/build cycle to fully take. It's especially
confusing because after a build cycle one ends up with a build partially using
the old and partially the new libraries.

I unfortunately don't see a localized fix for this. Afaict we'd need to change
all client build rules to also have a dependency on the library?

Greetings,

Andres Freund

[1] https://www.gnu.org/software/make/manual/html_node/Prerequisite-Types.html

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Arjan van de Ven 2021-10-26 18:13:14 Re: src/port/snprintf.c: Optimize the common base=10 case in fmtint
Previous Message Andrew Dunstan 2021-10-26 17:59:34 Re: pg_dump versus ancient server versions