Feedback-ID: id4a34324:Fastmail
Date: Mon, 30 Mar 2026 08:39:23 -0400
From: Andres Freund <andres@anarazel.de>
To: "Anton A. Melnikov" <a.melnikov@postgrespro.ru>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>
Subject: Re: .bc files build dependency issues.
Message-ID: <7ihq6kmoruqicq5qqns3wicb4yydrqgy4iqvke3wafa5f2h6i4@5y4pwfuogh2y>
References: <b0574669-10f1-44ec-93fe-e3a03b246005@postgrespro.ru>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <b0574669-10f1-44ec-93fe-e3a03b246005@postgrespro.ru>
Archived-At: 
 <https://www.postgresql.org/message-id/7ihq6kmoruqicq5qqns3wicb4yydrqgy4iqvke3wafa5f2h6i4%405y4pwfuogh2y>
Precedence: bulk

Hi,

On 2026-03-30 10:28:30 +0300, Anton A. Melnikov wrote:
> While the original discussion was about a specific commit
> ecaf7c5df and related fix in 16492df7 for 16+
> and its effects  [1], i would like to point out that
> a similar behavior can be observed
> on earlier stable branches (REL_14/15_STABLE) as well.

> Using a pre-built source tree and the following reproduction
> that is slightly modified version from [2] which suggests
> that this is a known pattern for exposing the issue:
> 
> cd src/backend/parser
> rm -rf .deps/ gram.c scan.c *.o *.bc gram.h
> make parser.bc

IOW, if you do something that intentionally breaks things, things break.


> result in:
> 
> $ make parser.bc
> /usr/lib/llvm-16/bin/clang -Wno-ignored-attributes -fno-strict-aliasing -fwrapv -fexcess-precision=standard -Xclang
> -no-opaque-pointers -Wno-unused-command-line-argument -Wno-compound-token-split-by-macro -Wno-deprecated-non-prototype
> -O2  -I. -I. -I../../../src/include -D_GNU_SOURCE  -I/usr/include/libxml2 -I/usr/local/include -flto=thin -emit-llvm
> -c -o parser.bc parser.c
> In file included from parser.c:25:
> ../../../src/include/parser/gramparse.h:29:10: fatal error: 'parser/gram.h' file not found
> #include "parser/gram.h"
>          ^~~~~~~~~~~~~~~
> 1 error generated.
> make: *** [../../../src/Makefile.global:1084: parser.bc] Error 1
> 
> Since this can be reproduced reliably on REL_14/15, it seems worth asking
> whether fix for .bc files (or something equivalent) should also be
> backpatched to lower branches, instead of only applying to newer ones.

This seems like a complete non-issue to me. What is the actual problem here?
You intentionally removed dependency information. You intentionally removed
part of the generated files, but not all of them.


> Also as a possible improvement perhaps make the fix less strictly?
> I.e. replace each .bc file's dependency on all .o files
> with a more weak dependency just on all source .c files.
> Namely use:
> 
> $(patsubst %.o,%.bc, $(OBJS)): $(patsubst %.o,%.c, $(OBJS))
> 
> instead of:
> 
> $(patsubst %.o,%.bc, $(OBJS)): $(OBJS)
> 
> This might still preserve correct build ordering while reducing
> unnecessary coupling between targets.

It'd also be broken if one of the source files is built from c++, for example.


> As a result make will run faster during parallel builds since it won't wait
> for absolutely all the object files to be created.
> 
> For example, rebuilding parser.bc would then require only
> a single compiler call in src/backend/parser instead of many.

The other file also needs to be built. The .bc->.o dependency will actually
often prevent redundant error messages.

Greetings,

Andres Freund