Details
-
Bug
-
Status: Confirmed (View Workflow)
-
Blocker
-
Resolution: Unresolved
-
11.4.5
-
Fedora41 + gcc (GCC) 14.2.1 20250110 (Red Hat 14.2.1-7)
Description
BUILD: mariadb-lts git_mariadb_11.4.4
on Fedora41 + gcc (GCC) 14.2.1 20250110 (Red Hat 14.2.1-7)
OK
-> log: https://pastebin.com/aHQzV8Qk
BUILD: mariadb-lts git_mariadb_11.4.5
on Fedora41 + gcc (GCC) 14.2.1 20250110 (Red Hat 14.2.1-7)
FAILS
-> https://pastebin.com/5FRnJnfy, line # 907
```
...
/builddir/build/BUILD/mariadb-lts-git_mariadb_11.4.5-build/server.git-e2c746b03313677dd880ed7f5ada49c17c8c1b1c/storage/innobase/include/trx0trx.h: In member function ‘void trx_t::mutex_lock()’:
/builddir/build/BUILD/mariadb-lts-git_mariadb_11.4.5-build/server.git-e2c746b03313677dd880ed7f5ada49c17c8c1b1c/storage/innobase/include/trx0trx.h:674:13: error: ‘mutex_owner’ was not declared in this scope
674 | assert(!mutex_owner.exchange(pthread_self(),
| ^~~~~~~~~~~ /builddir/build/BUILD/mariadb-lts-git_mariadb_11.4.5-build/server.git-e2c746b03313677dd880ed7f5ada49c17c8c1b1c/storage/innobase/include/trx0trx.h: In member function ‘void trx_t::mutex_unlock()’: /builddir/build/BUILD/mariadb-lts-git_mariadb_11.4.5-build/server.git-e2c746b03313677dd880ed7f5ada49c17c8c1b1c/storage/innobase/include/trx0trx.h:680:12: error: ‘mutex_owner’ was not declared in this scope 680 |
assert(mutex_owner.exchange(0, std::memory_order_relaxed) == |
| ^~~~~~~~~~~ ... ``` |
Attachments
Issue Links
- is duplicated by
-
-
- Closed
-
Activity
pastebin has decided the log's "harmful" and deleted it ...
reposting ->
BUILD: mariadb-lts git_mariadb_11.4.4
-> https://gist.github.com/pgnd/7973b0eb1f0d787c55a4c020c0fb00a6
A combination of
-DCMAKE_BUILD_TYPE=None '-DCMAKE_C_FLAGS=-DDBUG_OFF' '-DCMAKE_CXX_FLAGS=-DDBUG_OFF'
|
is what makes it different from our CI builds, both release and debug, and apparently with the change above it becomes too confusing for the build to get through.
-DCMAKE_BUILD_TYPE=RelWithDebInfo is fine; the absence of DCMAKE_BUILD_TYPE defaults to RelWithDebInfo; and -DCMAKE_BUILD_TYPE=Debug with simultaneous DBUG_OFF would fail regardless the recent change.
I think it's not the first time when -DCMAKE_BUILD_TYPE=None causes a trouble, maybe something should be done about it.
it's been quite awhile, so my specifics are fuzzy ...
iirc, i'd
```
- CMAKE_BUILD_TYPE="Release"
+ CMAKE_BUILD_TYPE="None"
```
as =Release was pulling in default flags that were causing me some grief with my local builds.
=None gave me a preferable blank-slate to work with.
pgnd None of the pastes work any more.
Can you please paste your cmake line? If you're still having trouble, it's likely because you are overwriting CMAKE_C_FLAGS and CMAKE_CXX_FLAGS instead of appending to them.
Similar/same issue was reported in gentoo with the full log, see https://bugs.gentoo.org/949720.
> your cmake line
the attachment, `cmake.txt`
WORKS for 11.4.4/release build
FAILS for 11.4.5/release build
WORKS for 11.6/branch build
I have bisected the issue between 11.4.4 and 11.4.5 to the commit 43233fe469fd ("Fix -DBUILD_CONFIG=mysql_release to keep standard cmake flags").
same as mentioned above, correct?
> Daniel Black added a comment - 2025-02-06 23:56
> regression from 43233fe469fd85d7c64978f01ed68de3288d43cc
regression from 43233fe469fd85d7c64978f01ed68de3288d43cc
mutex_owner is behind UNIV_DEBUG
diff --git a/storage/innobase/include/trx0trx.h b/storage/innobase/include/trx0trx.hindex 889437aa1d2..ef11f5edd1c 100644--- a/storage/innobase/include/trx0trx.h+++ b/storage/innobase/include/trx0trx.h@@ -658,10 +658,8 @@ struct trx_t : ilist_node<>/** mutex protecting state and some of lock(some are protected by lock_sys.latch) */srw_spin_mutex mutex;-#ifdef UNIV_DEBUG/** The owner of mutex (0 if none); protected by mutex */std::atomic<pthread_t> mutex_owner{0};-#endif /* UNIV_DEBUG */public:void mutex_init() { mutex.init(); }void mutex_destroy() { mutex.destroy(); }