On Mar 31 13:06, Corinna Vinschen wrote: > On Mar 31 17:18, Yuyi Wang wrote: > > > I tested this scenario, and this problem only occurs with > > > dlopening cygwin1.dll. > > > > Not only cygwin1.dll, but also native dlls, e.g., kernel32.dll or > > user32.dll. > > I haven't tested the next release, but do you think it's the same reason for > > win32 dlls? > > No, it's not. Native DLLs are not taken into account because they > don't call into Cygwin's dll_dllcrt0 on init, so they are not > added to the DLL list. > > Hmm. > > I'll have to check if we should add them to the dll list or not. > We certainly don't need them for atexit and stuff, but the dlopen > counting might be necessary at fork.
FTR, even if we keep track of native dlopen'ed DLLs, we can't reproduce their state after fork. We must not even try, because certain Windows DLLs choke on reproducing their data and bss segments and misbehave. So we can only keep track of the number of dlopen/dlclose calls for them. Corinna
