On 09/11/2025 09:02, Takashi Yano wrote:
Hi Mark,
On Sun, 9 Nov 2025 00:09:07 -0800
Mark Geisert wrote:
Hi Takashi,
On 11/4/2025 8:58 PM, Takashi Yano wrote:
On Tue, 28 Oct 2025 20:48:40 +0900
Takashi Yano wrote:
Takashi Yano (2):
Cygwin: dll_init: Call __cxa_finalize() for DLL_LOAD even in
exit_state
Cygwin: dll_init: Don't call dll::init() twice for DLL_LOAD.
winsup/cygwin/dll_init.cc | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)
--
2.51.0
Could anyone please review if these patches make sense?
The patches look fine to me. Do you happen to have an STC that
demonstrates to you the issue is fixed with your patch?
Thanks for reviewing. The STC is the attachment files in
https://cygwin.com/pipermail/cygwin/2025-October/258919.html
I'm finding it pretty hard to reason about what the possible
combinations that should be considered are.
Like, what is the spanning set? I guess we have:
1. A single DLL X, directly linked with by executable
2. A single DLL X, dlopened and dlclosed (subcases where it does this
during constructor/destructors and otherwise?)
3. As above, but X is directly linked with Y
4. As above, but X is dlopens/dlcloses Y
5. more???
If I understood all that, then maybe I'd have some suggestions about how
the comments can be written to explain why what it's doing is the right
thing in the various situations.
I guess it's possible to extend that STC to cover all those?
