|
Message-ID: <20190710224301.GZ1506@brightrain.aerifal.cx> Date: Wed, 10 Jul 2019 18:43:01 -0400 From: Rich Felker <dalias@...c.org> To: musl@...ts.openwall.com Subject: Re: open64 and similar On Wed, Jul 10, 2019 at 06:03:51PM -0400, Andrew Bell wrote: > musl does the following: > > #define open64 open > > This can cause an infinite loop for the following code: > > class Foo > { > public: > int open64() > { open(); } > }; > > Perhaps it would be better to supply open64 and have it call open, rather > than #define it? There are several other xxx64 functions also defined that > could cause problems with unfortunate code. The intent is that we're trying to prevent actual references to the legacy open64, etc. symbols while supporting code that's wrongly attempting to use them. At the time this was added, that was still a problem in lots of software; I don't know if it's since been fixed. Originally, the intent was that the symbols exist *only* as ABI, for ABI-compat loading of glibc-linked libs, and not as API that programs compiled against musl could use. However broken configure scripts checked for the symbol definition by linking a test program using a *fake* declaration of the symbol, without including the header, then wrongly picked up that it was available, and compiled wrong code later at compiel time due to implicit-function-declaration. So the macro redirections were added. I'd like it if we could remove this stuff entirely, except for the ABI-compat. Maybe it could be done by getting rid of the actual symbols and just putting magic in the dynamic linker to resolve them to the non-64 ones. Anyway, this is not the first time someone's hit a problem from it with C++, which is caused by GCC's unconditional (and wrong) definition of _GNU_SOURCE in C++ mode. So we really should try to find a reasonable fix... Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.