On Mon, 3 Mar 2025 at 09:10, Christophe Lyon christophe.lyon@linaro.org wrote:

Hi Jonathan,

On Sun, 2 Mar 2025 at 23:28, Jonathan Wakely via Gcc-regression gcc-regression@gcc.gnu.org wrote:

...

On Sun, 2 Mar 2025 at 02:42, ci_notify@linaro.org wrote:

...

Dear contributor,

Our automatic CI has detected problems related to your patch(es). Please find some details below.

In arm-eabi cortex-m7 hard, after: | commit gcc-15-7765-g3866ca796d5 | Author: Jonathan Wakely jwakely@redhat.com | Date: Thu Feb 27 13:27:17 2025 +0000 | | libstdc++: Fix ranges::move and ranges::move_backward to use iter_move [PR105609] | | The ranges::move and ranges::move_backward algorithms are supposed to | use ranges::iter_move(iter) instead of std::move(*iter), which matters | for an iterator type with an iter_move overload findable by ADL. | ... 16 lines of the commit log omitted.

Produces 8 regressions: | | regressions.sum: | Running libstdc++:libstdc++-dg/conformance.exp ... | FAIL: 25_algorithms/move/constrained.cc -std=gnu++20 (test for excess errors) | UNRESOLVED: 25_algorithms/move/constrained.cc -std=gnu++20 compilation failed to produce executable | FAIL: 25_algorithms/move/constrained.cc -std=gnu++26 (test for excess errors) | UNRESOLVED: 25_algorithms/move/constrained.cc -std=gnu++26 compilation failed to produce executable | ... and 4 more

It looks like thumb has some unusual linking requirements that I'm not familiar with, so undefined functions (which are never actually called in the test) cause linker errors:

/home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/x86_64-pc-linux-gnu/arm-eabi/gcc-gcc.git~master-stage2/arm-eabi/libstdc++-v3/include/bits/iterator_concepts.h:155:(.text._Z6test06N8pr1056091IE+0x6): undefined reference to `pr105609::iter_move(pr105609::I const&)' /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/destdir/x86_64-pc-linux-gnu/arm-eabi/bin/ld: (_ZN8pr1056099iter_moveERKNS_1IE): Unknown destination type (ARM/Thumb) in /tmp/ccqKQGzN.o

This is a "recent" linker warning, which I added sometime during last year (so you need recent binutils if you want to try to reproduce the problem). It means that for some reason the destination symbol lacks a tag (asm directive) saying whether it's a thumb or an arm function. This is used by the linker when deciding which type of stub to insert (if any). For instance: ".type myfunc, %function" but normally the compiler adds this for you.

But you mention "undefined functions", so that would probably be the reason, but how can the link succeed? Are you linking with "-z undefs" ?

See e.g. libstdc++-v3/testsuite/25_algorithms/move/constrained.cc The test06 function uses undefined member functions such as I::operator== but because test06 is never called, it shouldn't matter.

I can move test06 to a separate { dg-do compile } test, where the linker isn't involved at all.

Thanks,

Christophe

...

/home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/x86_64-pc-linux-gnu/arm-eabi/gcc-gcc.git~master-stage2/arm-eabi/libstdc++-v3/include/bits/iterator_concepts.h:155:(.text._Z6test06N8pr1056091IE+0x6): dangerous relocation: unsupported relocation /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/x86_64-pc-linux-gnu/arm-eabi/gcc-gcc.git~master-stage2/arm-eabi/libstdc++-v3/include/bits/ranges_algobase.h:326:(.text._Z6test06N8pr1056091IE+0xc): undefined reference to `pr105609::I::operator++()' /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/destdir/x86_64-pc-linux-gnu/arm-eabi/bin/ld: (_ZN8pr1056091IppEv): Unknown destination type (ARM/Thumb) in /tmp/ccqKQGzN.o /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/x86_64-pc-linux-gnu/arm-eabi/gcc-gcc.git~master-stage2/arm-eabi/libstdc++-v3/include/bits/ranges_algobase.h:326:(.text._Z6test06N8pr1056091IE+0xc): dangerous relocation: unsupported relocation /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/destdir/x86_64-pc-linux-gnu/arm-eabi/bin/ld: /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/x86_64-pc-linux-gnu/arm-eabi/gcc-gcc.git~master-stage2/arm-eabi/libstdc++-v3/include/bits/ranges_algobase.h:323:(.text._Z6test06N8pr1056091IE+0x16): undefined reference to `pr105609::I::operator==(pr105609::I) const' /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/destdir/x86_64-pc-linux-gnu/arm-eabi/bin/ld: (_ZNK8pr1056091IeqES0_): Unknown destination type (ARM/Thumb) in /tmp/ccqKQGzN.o /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/x86_64-pc-linux-gnu/arm-eabi/gcc-gcc.git~master-stage2/arm-eabi/libstdc++-v3/include/bits/ranges_algobase.h:323:(.text._Z6test06N8pr1056091IE+0x16): dangerous relocation: unsupported relocation

I think I should be able to workaround this.

On Wed, 5 Mar 2025 at 23:32, Jonathan Wakely jwakely@redhat.com wrote:

On Mon, 3 Mar 2025 at 09:27, Jonathan Wakely jwakely@redhat.com wrote:

...

On Mon, 3 Mar 2025 at 09:10, Christophe Lyon christophe.lyon@linaro.org wrote:

...

Hi Jonathan,

On Sun, 2 Mar 2025 at 23:28, Jonathan Wakely via Gcc-regression gcc-regression@gcc.gnu.org wrote:

...

On Sun, 2 Mar 2025 at 02:42, ci_notify@linaro.org wrote:

...

Dear contributor,

Our automatic CI has detected problems related to your patch(es). Please find some details below.

In arm-eabi cortex-m7 hard, after: | commit gcc-15-7765-g3866ca796d5 | Author: Jonathan Wakely jwakely@redhat.com | Date: Thu Feb 27 13:27:17 2025 +0000 | | libstdc++: Fix ranges::move and ranges::move_backward to use iter_move [PR105609] | | The ranges::move and ranges::move_backward algorithms are supposed to | use ranges::iter_move(iter) instead of std::move(*iter), which matters | for an iterator type with an iter_move overload findable by ADL. | ... 16 lines of the commit log omitted.

Produces 8 regressions: | | regressions.sum: | Running libstdc++:libstdc++-dg/conformance.exp ... | FAIL: 25_algorithms/move/constrained.cc -std=gnu++20 (test for excess errors) | UNRESOLVED: 25_algorithms/move/constrained.cc -std=gnu++20 compilation failed to produce executable | FAIL: 25_algorithms/move/constrained.cc -std=gnu++26 (test for excess errors) | UNRESOLVED: 25_algorithms/move/constrained.cc -std=gnu++26 compilation failed to produce executable | ... and 4 more

It looks like thumb has some unusual linking requirements that I'm not familiar with, so undefined functions (which are never actually called in the test) cause linker errors:

/home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/x86_64-pc-linux-gnu/arm-eabi/gcc-gcc.git~master-stage2/arm-eabi/libstdc++-v3/include/bits/iterator_concepts.h:155:(.text._Z6test06N8pr1056091IE+0x6): undefined reference to `pr105609::iter_move(pr105609::I const&)' /home/tcwg-buildslave/workspace/tcwg_gnu_1/abe/builds/destdir/x86_64-pc-linux-gnu/arm-eabi/bin/ld: (_ZN8pr1056099iter_moveERKNS_1IE): Unknown destination type (ARM/Thumb) in /tmp/ccqKQGzN.o

This is a "recent" linker warning, which I added sometime during last year (so you need recent binutils if you want to try to reproduce the problem). It means that for some reason the destination symbol lacks a tag (asm directive) saying whether it's a thumb or an arm function. This is used by the linker when deciding which type of stub to insert (if any). For instance: ".type myfunc, %function" but normally the compiler adds this for you.

But you mention "undefined functions", so that would probably be the reason, but how can the link succeed? Are you linking with "-z undefs" ?

See e.g. libstdc++-v3/testsuite/25_algorithms/move/constrained.cc The test06 function uses undefined member functions such as I::operator== but because test06 is never called, it shouldn't matter.

I can move test06 to a separate { dg-do compile } test, where the linker isn't involved at all.

I hope this is fixed at

r15-7842-gc21d5a3591fd761872e18278e1cd8ec18e36d4cb

Indeed, the tests now pass.

I still haven't understood why the original version passed on some targets? (I assume so, I suppose you tested the patch before committing ;-) ) Even if not called, test06 would still have references I::operator== whatever the target?

Thanks,

Christophe