added: cli option --dump-config prints loaded config and exits

[wolfgangwalther]

As discussed in #1659 (comment), this adds the CLI option --dump-config, which prints the config after loading and then exits. The printed config is itself a valid postgrest config file.

This is mostly useful as a base for future improvements. When adding aliases for some config options (#1668 (comment)) or reading options directly from environment variables (#1624, #1572), this can be used as the base to write proper tests.

Also added colors to our io-tests. They are all nice and green now. Much easier to spot the occasional red one...

[wolfgangwalther]

Lot's of haskell again, @steve-chavez. Looking forward to your feedback! I already learned a bit from hlint :D.

[wolfgangwalther]

Cirrus seems to be hitting a build timeout of 15 minutes now, saw that for a couple of commits in different PRs now. @steve-chavez you added a timeout of 30 minutes a little time ago. Did they lower the timeout or what's happening?

[steve-chavez]

Cirrus seems to be hitting a build timeout of 15 minutes now, saw that for a couple of commits in different PRs now

@wolfgangwalther Hmm.. are you sure it's a timeout? It looks like it's failing because of low memory: haskell/cabal#5505.

I saw https://cirrus-ci.com/task/6377627252424704, that shows:

cabal: Failed to build hasql-implicits-0.1.0.2 (which is required by
exe:postgrest from postgrest-7.0.1). The build process was killed (i.e.
SIGKILL). The typical reason for this is that there is not enough memory
available (e.g. the OS killed a process using lots of memory).

Maybe we should try adding -j1 to cirrus(I remember -j1 didn't work before, but might work now).

(I'll try fixing this on another PR)

Edit: The-j1 seems to have worked. I added the change on #1600 (was also getting the cirrus error here).

[wolfgangwalther]

@wolfgangwalther Hmm.. are you sure it's a timeout? It looks like it's failing because of low memory: haskell/cabal#5505.

Absolutely, the low memory makes sense. I did not look carefully - I just figured it was not a real build error. I had 3 cirrus builds failing at 15 minutes, so I was a bit too quick to assume it was a timeout thing.

Edit: The-j1 seems to have worked. I added the change on #1600 (was also getting the cirrus error here).

Sweet, so I won't look at this here and just see green where red is :)

[steve-chavez]

The printed config is itself a valid postgrest config file.

I thought this would mean I could do postgrest --dump-config > pgrst.conf and generate a default config. But now I get it's mostly for tests. I'll give it a review now :)

[wolfgangwalther]

I thought this would mean I could do postgrest --dump-config > pgrst.conf and generate a default config. But now I get it's mostly for tests.

I think we're missing 3 things to be able to do that:

• the FILENAME argument must be optional. This is something that I have to change anyway to support "all-environment-only" configurations.
• with --dump-config enabled, postgrest should not throw when missing one of the required settings. This is something that could happen with the environment variables approach - not 100% yet. Depends on implementation of that.
• A good example config file needs comments, just like we provide them with the help output.

Maybe we can do that a bit different - what if we split the output of --help / the error output on stdout and stderr, so that only the example config file is on stdout and everything else on stderr? Then you should be able to do postgrest --help > example.config and have our example ready to go.

[steve-chavez]

what if we split the output of --help / the error output on stdout and stderr, so that only the example config file is on stdout and everything else on stderr? Then you should be able to do postgrest --help > example.config and have our example ready to go.

That sounds good! Seems like a simpler implementation.

[wolfgangwalther]

what if we split the output of --help / the error output on stdout and stderr, so that only the example config file is on stdout and everything else on stderr? Then you should be able to do postgrest --help > example.config and have our example ready to go.

That sounds good! Seems like a simpler implementation.

Had a look at that.. and the implementation is not as easy as thought. The whole thing is dumped by Options.Applicative.

Implementation-wise we could just do the following:

• remove the example config from the options help output
• add another CLI flag --dump-example-config to print just that

I think that's also better to use. WDYT?

---

Anyway, this would be something for a different PR. I addressed the other comments, CI is passing -> merge time!