macros.meson: use setup --reconfigure for in-place builds #13340
Conversation
|
This slows down builds as we will now always reconfigure whereas we previously didn't if no options change. I think meson should be fixed to not log a message if options are passed to meson setup. Also --reconfigure doesn't actually do a proper job anyway as environment variables are not reread so now flags will not apply anyway. |
What do you mean? We should silently do nothing instead of loudly doing nothing?
Indeed. If you want to reapply all project setup then you have to wipe the build directory and possibly use ccache. |
Right now we loudly log and then do something. It makes users think that they're doing something wrong even though running meson setup on an already configured build directory is supported now. So I don't see the point in complaining about users doing it tbh. |
|
Hmm, indeed. I misunderstood what $ meson configure build|grep remote
remote disabled [enabled, disabled, auto] support for "journal over the network"
$ meson setup build -Dremote=enabled
Directory already configured.
Just run your build command (e.g. ninja) and Meson will regenerate as necessary.
Run "meson setup --reconfigure to force Meson to regenerate.
If build failures persist, run "meson setup --wipe" to rebuild from scratch
using the same options as passed when configuring the build.
$ meson configure build|grep remote
remote enabled [enabled, disabled, auto] support for "journal over the network" So it looks like a refusal, but actually the changes are applied. So yeah, I agree with Daan that the best option would be drop this verbose message. Instead, |
|
Updated to just print one-line message and continue. |
| print('Directory already configured, applying new settings.') | ||
| else: | ||
| print('Directory already configured.') |
There was a problem hiding this comment.
Why drop the mconf.run_impl() call? Shouldn't this change be just about changing the logging but keeping the implementation the same?
|
Why drop the `mconf.run_impl()` call? Shouldn't this change be just about changing the logging but keeping the implementation the same?
That code was there because the function was trying to exit early.
By _not_ doing that, we do the right thing, and the configuration options
are applied later. I'll reword the commit message to explain this better.
|
Current behaviour was confusing. It *looked* like meson refused the
operation, but in fact it was applying the configuration changes:
Example from systemd repo:
$ meson configure build|grep remote
remote disabled [enabled, disabled, auto] support for "journal over the network"
$ meson setup build -Dremote=enabled
Directory already configured.
Just run your build command (e.g. ninja) and Meson will regenerate as necessary.
Run "meson setup --reconfigure to force Meson to regenerate.
If build failures persist, run "meson setup --wipe" to rebuild from scratch
using the same options as passed when configuring the build.
$ meson configure build|grep remote
remote enabled [enabled, disabled, auto] support for "journal over the network"
mconf.run_impl() sets the options and returns 0. By removing the special
code that was shoehorned into validate_dirs(), we let the build continue.
We now print a short message to explain what is happening.
This fixes the unexpected behaviour where the configuration options would
be changed, but the setup wouldn't be done. There is no good reason for
'mkosi setup' to not set up the directory. Either it should refuse completely
or work fully. The latter seems more useful and is closer to current behaviour,
so let's do that.
(Inspired by systemd/systemd#32321.)
The case where the build directory already exists is rare, usually rpm builds are done from scratch. But in case of in-place builds (where the rpm package is built from an already-unpacked development tree) and in case of short-circuit builds where parts of the build are skipped to save time, the directory may exist. Obviously, such solution are not perfect, the build is not done in a pristine environment and the user has to be careful, but users already know this and are willing to take the risk because of the massive time savings.
Without --reconfigure, 'meson setup' says "directory was already configured" and proceeds. Add --reconfigure so that the configuration is applied. If the build directory doesn't exist, this has no effect.
(I initially wanted to do this in systemd, but it seems more reasonable to push it to the macros, since it applies to all rpm packages: systemd/systemd#32321)