Building M4-1.4.19 consistently fails: gl_xlist.h issues half a dozen warnings about converting integers to pointers. In gl_list.h these are explicitly declared to be pointers to structs. I don't know why g++ treats them as integers and then treats warnings as an error.

I've set CXXFLAGS to "-std=gnu++2b -Wno-error=int-conversion" and sometimes "-w" but g++ seems to ignore these flags.

would be nice to see some console output or similar, otherwise hard to say anything. g++ cannot ignore those flags "just for fun"

Hi pan64, I've attached config.log and the environment details but LQ apparently doesn't allow .h attachments. The files in question are gl_list.h, gl_xlist.h and gl_oset.h. I could rename them but perhaps you have a tarball for m4-1.4.19.

An error occurs first in gl_oset.h where there is a _GL_ATTRIBUTE_NODISCARD on line 275. I moved it to a separate line, tried that, added a ; tried that and finally took it out. Don't know whether that causes the susequent warnings in gl_xlist.h about pointer conversion. I think "-Wint-conversion" is an error in C++ and so it compiles only to the end of the file.

It seems that the environment g++ flags are overruled by the m4 flags which may include -Werror somewhere. If you want to see the original error in gl_oset.h I can create another config.log for that by re-inserting _GL_ATTRIBUTE_NODISCARD before int in line 275.

Thx for interest.

PS I'm glad I did have to make an error log. It shows that these attribute errors originate in gl_list.h, included later by gl_oset.h; better try to solve them in gl_list.h I suppose.

Attached Files

	make0.log (22.5 KB, 3 views)
	config0.log (61.1 KB, 3 views)
	env.txt (528 Bytes, 3 views)
	make1.log (714 Bytes, 3 views)

I see that another syntax for attributes puts the attribute after the closing } of the definition so I did that with all the "errors" in gl_list.h and the one in gl_oset.h and then it built ok. I hesitate to mark this solved as every change elicited a warning "attribute nodiscard ignored" in the error log (attached). But i'm proceeding to "ncurses". I'll mark it solved if pan64 says ok.

Attached Files

make4.log (9.8 KB, 4 views)

Put a .txt suffix on your header files and LQ will accept them.

JohnHendry, as I'm looking at the M4 source code (and your TXT outputs) it's mostly C and H files giving you the warnings, so maybe you want to set `CFLAGS` in addition to `CXXFLAGS`?

Also running `./configure --help` hints at the `--enable-c++` option:

Yes, indeed, Projectile, I realised that it was mostly C after a while and I put in my CFLAGS -Wno-error=attributes and -Wno-error=int-conversion, without effect however. But I hadn't seen --enable-c++. The LFS-11.3 book tends to minimise options till the final installation of the package a bit later in the procedure. And thx, Hazel, I could rename these .h files, just didn't want to offend LQ. Anyway I found a workaround for the attributes and if m4 can manage without them for two more chapters I'll install it again.

JohnHendry, I don't think `CFLAGS` and `CXXFLAGS` would simply get ignored: in my generated Makefile they appear at lines 394 and 401 respectively.

Additionally, on Slackware64 15.0, GCC 11.2.0, M4 1.4.19 builds without any errors. Perhaps this is a bug in GCC 12, or specifically in your LFS compiler?

Anyway, glad you found a workaround.

You're right, Projectile, they are where you said; another line says CXX = no, however, presumably because the command-line only specified build, host and target, and another line says CFLAG_VISIBILITY= . My workaround was to move the offending "_GL_ATTRIBUTE_NODISCARD" phrases after the definition } where they still form part of it, according to what I read. But part of it or not the compiler says "ignoring attribute nodiscard" for every one so there's probably a better way. At least it builds.