[MDEV-23230] wsrep files installed when built without WSREP Created: 2020-07-20 Updated: 2023-02-13 Resolved: 2022-11-28 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | Compiling |
| Affects Version/s: | 10.4.13, 10.5.4, 10.5.5, 10.5.8, 10.5.16, 10.6.11 |
| Fix Version/s: | 10.11.2, 10.5.19, 10.6.12, 10.7.8, 10.8.7, 10.9.5, 10.10.3 |
| Type: | Bug | Priority: | Major |
| Reporter: | Chris Ross | Assignee: | Andrew Hutchings |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Environment: |
Linux |
||
| Attachments: |
|
||||||||
| Issue Links: |
|
||||||||
| Description |
|
When building with -DWITH_WSREP=OFF -DPLUGIN_WSREP_INFO=NO, I am still finding many wsrep-related files installed in the INSTALLDIR. Specifically:
Whether or not the includes files should be installed I don't know, and the SQL in wsrep_notify and the README-wsrep could also be argued. But, certainly the man pages should not be installed as the tools are not built or installed. I would suggest that none of these should be installed, and that the build configurtion should turn them all off if -DWITH_WSREP=OFF. |
| Comments |
| Comment by Chris Ross [ 2020-07-31 ] | |||||||||||||
|
The following patch will prevent installation of the man pages and support/doc files. Removing the headers would be a little more complicated, and I'm not 100% sure those aren't something some users might want even when building without wsrep. Insight would be appreciated, I can figure out how to prevent those as well. This is a quick patch, style may not be what you want, but it accomplishes my goal. | |||||||||||||
| Comment by Chris Ross [ 2021-06-28 ] | |||||||||||||
|
This issue still affects 10.5.10, and is causing undesired files to be installed without us manually patching this. could you please integrate the supplied change or one like it? | |||||||||||||
| Comment by Chris Ross [ 2022-07-07 ] | |||||||||||||
|
This is still affecting 10.5.13 and 10.5.16. We're continuing to patch our local build process to avoid these unnecessary files being installed on our product. | |||||||||||||
| Comment by Daniel Black [ 2022-11-16 ] | |||||||||||||
|
cross, your on the right track. Small style bits to group just the WSREP component different:
10.6 starts to group some WSREP components under variables so taking some of those changes limited to WSREP, back porting those to 10.5, and apply your changes and that would make the merge up a little easier. Patch/pull request welcome. If attaching patch here can you indicate the licence choice: BSD New 3 Clause License, or under the MariaDB Contributor Agreement. | |||||||||||||
| Comment by Chris Ross [ 2022-11-16 ] | |||||||||||||
|
I took your suggested structure, and the only place a saw a change for variables was in man/CMakeLists.txt, and I pulled that in. Also I noted that in 10.6 there is still a wsrep man page outside of that variable. Should wsrep_sst_mysqldump.1 be left out of MAN1_WSREP? I presume it was an oversight and have corrected in my patch. | |||||||||||||
| Comment by Daniel Black [ 2022-11-17 ] | |||||||||||||
|
Thanks Chris, good corrections. Also excluded include/*/wsrep.h too and galera system files. https://github.com/MariaDB/server/pull/2334 for final review. | |||||||||||||
| Comment by Chris Ross [ 2022-11-17 ] | |||||||||||||
|
Thank you. I didn't catch that galera should be excluded too, thanks. And re STRING(APPEND, I knew about the "only works in recent cmake" but thought that the large number of other STRING() uses meant it was okay. Now I know better. | |||||||||||||
| Comment by Daniel Black [ 2022-11-17 ] | |||||||||||||
|
It was deceptive, all the STRING(APPEND was under Windows which has a higher CMake version threshold. We've still got cmake-ancient in rhel/cento-7. Glad your happy with it. Sorry it too so long. And thanks for the ping on maria-discuss. | |||||||||||||
| Comment by Chris Ross [ 2022-11-18 ] | |||||||||||||
|
Just confirmed that these files (or a very similar set) are also installed on 10.6.11 compiled with -DWITH_WSREP=OFF -DPLUGIN_WSREP_INFO=NO. Thanks. Looking forward to the fix getting into the trees. | |||||||||||||
| Comment by Chris Ross [ 2022-11-28 ] | |||||||||||||
|
Thanks. I see this got pulled into 10.5 today, but it is noted as fixed in 10.5.20, not 10.5.19 which should be the next out. So curious for a "why" on that, and a pointer to track this getting up into 10.6 (and later) so I can plan for which 10.6 release will have it. I hope/expect to have our internal product update to 10.6.x shortly. Thank you. | |||||||||||||
| Comment by Sergei Golubchik [ 2022-11-28 ] | |||||||||||||
|
must've been a mistake. Thanks | |||||||||||||
| Comment by Robin Newhouse [ 2023-01-05 ] | |||||||||||||
|
The changes introduce failures in MTR tests when WITH_WSREP=OFF. See |