Hi Pali, Hmm, it doesn't seem to be related, at least on first glance. I'm working on adding native TLS to gcc, but right now it uses software emulated TLS, which is what seems to be broken in that bug. I unfortunately doubt I am competent enough to fix that bug, however :(
best regards, Julian On Sat, Sep 14, 2024 at 5:37 PM Pali Rohár <[email protected]> wrote: > > Hello Julian, could you check if the TLS problem on which you working is > in binutils is related to this newly reported mingw-w64 bug? > https://sourceforge.net/p/mingw-w64/bugs/994/ > > On Monday 09 September 2024 11:03:47 Pali Rohár wrote: > > Hello Julian, it would be nice if you could take this feature. > > > > On Monday 09 September 2024 16:44:55 Julian Waters wrote: > > > I'm working on gcc and binutils at the moment to add Thread Local Storage > > > to both, if you all don't mind I'll also add the importing local symbol > > > feature to my bucket list as well once I'm done with that > > > > > > best regards, > > > Julian > > > > > > On Mon, Sep 9, 2024 at 4:40 PM Pali Rohár <[email protected]> wrote: > > > > > > > On Monday 09 September 2024 09:40:41 Jacek Caban wrote: > > > > > On 7.09.2024 00:00, Pali Rohár wrote: > > > > > > So maybe the imp symbol could be included only for gcc/binutils > > > > > > builds? > > > > > > > > > > > > > > > I think that we don't want to introduce differences like that. > > > > > Ideally, > > > > we > > > > > should be able to mix tools with crt builds with another tools (so, > > > > > for > > > > > example, use crt built with GCC with clang+lld and the other way > > > > > around). > > > > > > > > Martin in other email pointed that there are such differences due to > > > > llvm's dlltool is different from binutils's dlltool (usage of short vs > > > > long imports and non-usage of weak symbols). So as I understood, mixing > > > > is currently not possible. > > > > > > > > But I agree with you that mixing could be nice feature to have. > > > > > > > > > > > > > > As I mentioned, I think that it's better to prevent people from adding > > > > bogus > > > > > dllimport in the future code than silently accepting it. There is a > > > > concern > > > > > that such code may already exist and if it does, we shouldn't break > > > > > it. > > > > But > > > > > the fact that UCRT was fine without the symbol for years suggests that > > > > it's > > > > > not a problem in practice. > > > > > > > > > > > > > > > Jacek > > > > > > > > As I said, feel free to drop this imp symbol completely. It was just a > > > > my idea that it could be useful. But I fully understand your concern > > > > that it was not there before and seems that it is not needed in > > > > practice. > > > > > > > > I agree with you that the best would be to teach binutils's ld to do > > > > conversion with warning, like it is in msvc or clang/llvm. But this is > > > > long-term... > > > > > > > > > > > > _______________________________________________ > > > > Mingw-w64-public mailing list > > > > [email protected] > > > > https://lists.sourceforge.net/lists/listinfo/mingw-w64-public > > > > _______________________________________________ Mingw-w64-public mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
