Re: Refactoring the regression tests for more independence
Hi Tom, > FWIW, I tried to replicate this locally on my own RPi3B+, using > current Ubuntu 20.04.4 LTS (GNU/Linux 5.4.0-1066-raspi aarch64). > No luck: it all works fine for me. We have at least one Raspbian > buildfarm animal too, and it's not been unhappy either. I suspect > there is something odd about your environment settings. Thanks for sharing this. I repeated the experiment in a clean environment (Raspbian installed from scratch on a brand new SD-card) and can confirm that the problem is gone. Sorry for the disturbance. -- Best regards, Aleksander Alekseev
Re: Refactoring the regression tests for more independence
Aleksander Alekseev writes: > Sorry about the confusion regarding the environment differences. GCC > version is 8.3.0, Perl 5.28.1. All in all this is pretty much the > default Raspbian 10 environment, something you would typically get > after setting up your RPi 3 B+ using Raspberry Pi Imager and running > `apt update`, nothing exotic. Please let me know if there are any > other details of interest. FWIW, I tried to replicate this locally on my own RPi3B+, using current Ubuntu 20.04.4 LTS (GNU/Linux 5.4.0-1066-raspi aarch64). No luck: it all works fine for me. We have at least one Raspbian buildfarm animal too, and it's not been unhappy either. I suspect there is something odd about your environment settings. regards, tom lane
Re: Refactoring the regression tests for more independence
Hi Tom, > Since you haven't explained what's different about this environment, > it's hard to comment on these results. But is this really a stock > Postgres source tree, with no local modifications? The fragment of > src/test/regress/expected/copy.out that you show does not look > current. Yes, this is a stock PostgreSQL source code without any modification, with `git clean -dfx` etc. The fragment of copy.out probably doesn't look current because I was using `git bisect` and I'm on cc50080a82 right now. However the same tests fail on both `master` and `REL_15_STABLE`. It takes a while on Raspberry Pi to rebuild Postgres :) To clarify, the step that is failing is `./quick-build.sh`, or `make check' in this script to be precise. So postgresql.conf I'm using in single-install.sh has nothing to do with the problem, this step is not reached. Sorry about the confusion regarding the environment differences. GCC version is 8.3.0, Perl 5.28.1. All in all this is pretty much the default Raspbian 10 environment, something you would typically get after setting up your RPi 3 B+ using Raspberry Pi Imager and running `apt update`, nothing exotic. Please let me know if there are any other details of interest. I'll continue looking for the source of the problem and will post an update as soon as I have one. -- Best regards, Aleksander Alekseev
Re: Refactoring the regression tests for more independence
Aleksander Alekseev writes: > I wanted to test one of the patches we have for the July CF on the > Raspberry Pi 3 Model B+. It runs Raspbian GNU/Linux 10 (buster) and > Linux Kernel 5.10.60-v7+. > I discovered that the PostgreSQL tests don't pass in this environment. Since you haven't explained what's different about this environment, it's hard to comment on these results. But is this really a stock Postgres source tree, with no local modifications? The fragment of src/test/regress/expected/copy.out that you show does not look current. regards, tom lane
Re: Refactoring the regression tests for more independence
Julien Rouhaud writes: > On Mon, Feb 07, 2022 at 02:00:25PM -0500, Tom Lane wrote: >> Not too surprisingly, these patches broke during the commitfest. >> Here's a rebased version. >> I'm not sure that anyone wants to review these in detail ... >> should I just go ahead and push them? > I don't see anything shocking after a quick glance, and I don't think any > review is going to give any more confidence compared to the script-dep-testing > script, so +1 for pushing them since the cf bot is green again. Done, will watch the farm. regards, tom lane
Re: Refactoring the regression tests for more independence
On Mon, Feb 07, 2022 at 02:00:25PM -0500, Tom Lane wrote: > Not too surprisingly, these patches broke during the commitfest. > Here's a rebased version. > > I'm not sure that anyone wants to review these in detail ... > should I just go ahead and push them? I don't see anything shocking after a quick glance, and I don't think any review is going to give any more confidence compared to the script-dep-testing script, so +1 for pushing them since the cf bot is green again.
Re: Refactoring the regression tests for more independence
On Fri, Dec 24, 2021 at 05:00:17PM -0500, Tom Lane wrote: > While I've not done so here, I'm tempted to rename > create_function_0 to create_function_c and create_function_3 to > create_function_sql, to give them better-defined charters and > eliminate the confusion with trailing digits for variant files. +1 > (Maybe we should provide a way to run specified test script(s) *without* > invoking the whole schedule first.) +1 ; it can be done later, though. It's nice to be able to get feedback within a few seconds. That supports the idea of writing tests earlier. I guess this may expose some instabilities due to timing of autovacuum (which I'd say is a good thing). If you rearrange the creation of objects, that may provide an opportunity to rename some tables with too-short names, since backpatching would already have conflicts. -- Justin
