On 20/11/2021 20:03, Kevin Morris via pacman-dev wrote:
If a transaction exists, release it if possible. If `alpm_trans_release` errors out (returns a non-zero), return a -1 in `alpm_release`.
This patch addresses the following issue observed in `pyalpm`: https://gitlab.archlinux.org/archlinux/pyalpm/-/issues/25, where `pyalpm` ends up raising twice due to an unexpected failure during handle release, while a transaction is locked.
This is a modified version of https://patchwork.archlinux.org/project/pacman/patch/20200119191833.581492-1... and is a derivation of his (Morganamilo's) work.
Signed-off-by: Kevin Morris <kevr@0cost.org> --- lib/libalpm/alpm.c | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/lib/libalpm/alpm.c b/lib/libalpm/alpm.c index 6708e202..47d02fea 100644 --- a/lib/libalpm/alpm.c +++ b/lib/libalpm/alpm.c @@ -100,6 +100,10 @@ int SYMEXPORT alpm_release(alpm_handle_t *myhandle)
CHECK_HANDLE(myhandle, return -1);
+ if (myhandle->trans != NULL && alpm_trans_release(myhandle)) { + return -1; + } + /* close local database */ db = myhandle->db_local; if(db) {
So looking through the code, alpm_release calls many functions that may fail but only actually looks at the return value of alpm_unregister_all_syncdbs. alpm_unregister_all_syncdbs can only fail if there's a transaction active, which this patch prevents. We should probably error check all the function calls.