Ok, I just confirmed that the PR https://github.com/emscripten-core/emscripten/pull/17866 fixes the original issue with the spine-c runtime. Many thanks!
On Wednesday, 7 September 2022 at 10:46:45 UTC+2 Floh wrote: > Ok, ticket is here: > > https://github.com/emscripten-core/emscripten/issues/17808 > > Let me know if I can help with anything else :) > > On Tuesday, 6 September 2022 at 20:30:55 UTC+2 [email protected] wrote: > >> Yes, please file a ticket. >> >> I think the solution is going to be to set `force_object_files = True` on >> `libmalloc` in `tools/system_libs.spy`.. which means it (like compiler-rt >> et al) will be compiled as normal object files and not take part in LTO. >> >> I am curious why this doesn't happen for dlmalloc too though.. >> >> cheers, >> sam >> >> On Tue, Sep 6, 2022 at 10:19 AM Floh <[email protected]> wrote: >> >>> Ok, I could cobble together a surprisingly simple reproducer here: >>> >>> https://github.com/floooh/emsc-calloc-emmalloc-repro >>> >>> -flto is indeed needed. So the 4 ingredients are: >>> >>> 1. malloc+memset being 'optimized' to a call to calloc via -O1 or better >>> 2. ...where malloc is called through a function pointer >>> 3. ...and emmalloc is used instead of the vanilla allocator >>> 4. ...and -flto must be enabled >>> >>> ...if any of this is missing, it works :D >>> >>> Should I write an Emscripten ticket too? >>> On Tuesday, 6 September 2022 at 18:43:52 UTC+2 Floh wrote: >>> >>>> Part of the mystery can be solved because clang replaces malloc+memset >>>> pairs with a single calloc call (only when optimization is enabled): >>>> >>>> https://www.godbolt.org/z/8r3Ydjs6a >>>> >>>> ...which explains why the problem shows up in release mode, but not >>>> debug mode. >>>> >>>> There still must be something else going on though, because I'm using >>>> the malloc+memset combo all the time together with emmalloc (and I guess >>>> the missing piece is that the malloc call is going through the function >>>> pointer). >>>> >>>> I'll tinker around a bit more with compiler options (but since this is >>>> a library, I cannot dictate what compiler options the code is built with). >>>> >>>> Cheers! >>>> >>>> On Monday, 5 September 2022 at 19:31:17 UTC+2 [email protected] wrote: >>>> >>>>> The other thing this could be related to is LTO. IIUC LTO itself can >>>>> generate calls to buildin functions (such as malloc and calloc) that are >>>>> not present in the original program. Does disabling LTO fix the problem? >>>>> >>>>> On Mon, Sep 5, 2022 at 10:28 AM Sam Clegg <[email protected]> wrote: >>>>> >>>>>> Can you trying building with `-Wl,--trace-symbol=calloc` rather than >>>>>> `LLD_REPORT_UNDEFINED=1` (I guess there are still some issues with that >>>>>> setting). >>>>>> >>>>>> Does the link time error tell you why calloc is being pulled in? >>>>>> (The JS compiler linker errors should do this but they often just say >>>>>> "top >>>>>> level C/C++"). >>>>>> >>>>>> On Sat, Sep 3, 2022 at 6:03 AM Floh <[email protected]> wrote: >>>>>> >>>>>>> I just stumbled over a very weird problem in a new sample for my >>>>>>> sokol headers which includes a non-trivial 3rd party C library - the >>>>>>> spine-c runtime ( >>>>>>> https://github.com/EsotericSoftware/spine-runtimes/tree/4.1/spine-c/spine-c >>>>>>> ) >>>>>>> >>>>>>> When building in release mode (but not in debug mode) the linker >>>>>>> complains about an unresolved 'calloc' call. Adding detailed output >>>>>>> via LLD_REPORT_UNDEFINED=1 I get tons of: >>>>>>> >>>>>>> wasm-ld: error: >>>>>>> /Users/floh/projects/fips-sdks/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/lto/libemmalloc.a(emmalloc.o): >>>>>>> >>>>>>> undefined symbol: calloc >>>>>>> >>>>>>> ... how does that even make sense when emmalloc is supposed to >>>>>>> provide an implementation of calloc... anyway... >>>>>>> >>>>>>> The error goes away when compiling in debug mode (some differences >>>>>>> that come to mind to the release mode is that debug mode doesn't use >>>>>>> -flto, >>>>>>> and also doesn't run the closure compiler). >>>>>>> >>>>>>> The problem also goes away when not building with emmalloc. >>>>>>> >>>>>>> Now the weird thing is that the entire code this sample is built >>>>>>> from doesn't appear to call calloc anywhere... and that other samples >>>>>>> which >>>>>>> actually use calloc build just fines. >>>>>>> >>>>>>> Has anybody seen a similar problem yet, before I jump myself into >>>>>>> the rabbit hole? :) >>>>>>> >>>>>>> The problematic command line is: >>>>>>> >>>>>>> em++ -s DISABLE_EXCEPTION_CATCHING=1 -fno-exceptions -fno-rtti >>>>>>> -fstrict-aliasing -Wall -Wno-multichar -Wextra -Wno-unknown-pragmas >>>>>>> -Wno-ignored-qualifiers -Wno-long-long -Wno-overloaded-virtual >>>>>>> -Wno-deprecated-writable-strings -Wno-unused-volatile-lvalue >>>>>>> -Wno-inconsistent-missing-override -Wno-warn-absolute-paths >>>>>>> -Wno-expansion-to-defined -flto -O3 -DNDEBUG -s >>>>>>> DISABLE_EXCEPTION_CATCHING=1 --memory-init-file 0 -s >>>>>>> INITIAL_MEMORY=33554432 -s ERROR_ON_UNDEFINED_SYMBOLS=1 -s >>>>>>> NO_EXIT_RUNTIME=1 -s LLD_REPORT_UNDEFINED=1 -s ALLOW_MEMORY_GROWTH=1 -s >>>>>>> USE_WEBGL2=1 -s "MALLOC='emmalloc'" -s NO_FILESYSTEM=1 -s WASM=1 >>>>>>> --shell-file /Users/floh/projects/sokol-samples/webpage/shell.html -O3 >>>>>>> -flto --closure 1 -s ASSERTIONS=0 >>>>>>> sapp/CMakeFiles/spine-sapp.dir/spine-sapp.c.obj >>>>>>> sapp/CMakeFiles/spine-sapp.dir/spine-assets.c.obj -o >>>>>>> /Users/floh/projects/fips-deploy/sokol-samples/sapp-webgl2-wasm-vscode-release/spine-sapp.html >>>>>>> >>>>>>> libs/sokol/libsokol.a libs/spine-c/libspine-c.a libs/stb/libstb.a >>>>>>> libs/util/libfileutil.a fips-cimgui_cimgui/libcimgui.a >>>>>>> >>>>>>> -- >>>>>>> You received this message because you are subscribed to the Google >>>>>>> Groups "emscripten-discuss" group. >>>>>>> To unsubscribe from this group and stop receiving emails from it, >>>>>>> send an email to [email protected]. >>>>>>> To view this discussion on the web visit >>>>>>> https://groups.google.com/d/msgid/emscripten-discuss/c47cee25-6e4b-4fbe-9f6f-0d7da7e02562n%40googlegroups.com >>>>>>> >>>>>>> <https://groups.google.com/d/msgid/emscripten-discuss/c47cee25-6e4b-4fbe-9f6f-0d7da7e02562n%40googlegroups.com?utm_medium=email&utm_source=footer> >>>>>>> . >>>>>>> >>>>>> -- >>> You received this message because you are subscribed to the Google >>> Groups "emscripten-discuss" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to [email protected]. >>> >> To view this discussion on the web visit >>> https://groups.google.com/d/msgid/emscripten-discuss/f761c696-9943-474d-8bd8-6bccb0667f37n%40googlegroups.com >>> >>> <https://groups.google.com/d/msgid/emscripten-discuss/f761c696-9943-474d-8bd8-6bccb0667f37n%40googlegroups.com?utm_medium=email&utm_source=footer> >>> . >>> >> -- You received this message because you are subscribed to the Google Groups "emscripten-discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/emscripten-discuss/4fea7bd5-6a73-4d03-b481-37c28ea8c08fn%40googlegroups.com.
