I'm not sure, I don't think we want to call this a BC break (or tag it as such) to give the wrong impression that command output is considered within the BC promise.

Let's also friendly ping @Seldaek who implemented the existing logic of showing errors during quiet verbosity. Do you remember a use-case for this? (something with Composer maybe?) Is there something we should take in mind before merging this PR?

IMO it is better to keep displaying it at least if you have a ConsoleOutputInterface where you can output on stderr. Any machine readable output would generally be expected on stdout. By default it will output to stderr so it won't disturb anything.

So I personally would close this, because yes when running quiet in some script/CI and something major breaks and you do not see it at all, it sucks. But that's just my view of course.

Note: I see the point about logging duplicating things, that's not great indeed. I am not sure how else this could be mitigated. I don't really see a way to safely do this, and IMO duplicating uncaught exception sure not great but still better as not seeing them at all.

I get your point about stderr, however both stderr and stdout of processes are redirected to a container's output. So in hosting relying on containers, this is still an issue.

There are utils/techniques you can apply to hide the output of a command, but still get errors (based on exit codes). E.g. in the past, I've used the little chronic bash utility for this in our CI. And you can also use bash one-liners like output=`mycommand 2>&1` || echo $output to accomplish this.

Ok, then I guess all I can ask is to maybe make this optional so that apps that are pure CLI apps like Composer and not 12-factor style symfony envs with full logger config could maybe retain the quiet-level output?

Let's resume this PR?

Yes please :)

We need to make a decision (@symfony/mergers):

A) Make --quiet completely silent, also for errors (the current state of this PR).
This is a bit troubling for applications currently relying on seeing errors when running with quiet verbosity.
It however is much more compatible with 99% of the CLI tools in the world. Only getting output on error is still possible using the techniques I shared before.

B) Add a new verbosity (e.g. SHELL_VERBOSITY=-2) that disables error output.
No change of existing behavior, but introduces more work (and also requires discoverability) to get truly silent output like you get with other CLI tools.
Optionally, we might add some code in the bin/console recipe to default to -2 when using --quiet to remove this extra barrier for upgraded applications.

@wouterj using 2 > &1 and then complaining that it screws the processing of the output is void, as the screwing comes from the 2 > &1, not from using stderr propertly in the first place.

Thus, cases that want to process the command output programmatically are probably not using --quiet anyway (as there would be not output to process)

@wouter I looked in other projects and some commands use special options for this:

• Google Cloud command has --quiet to hide interactive messages and --no-user-output-enabled to suppress printing anything in stdout or stderr (https://cloud.google.com/sdk/gcloud#suppressing_prompting_writing_to_the_terminal_and_logging)
• AWS commands have --quiet to not output information and --only-show-errors to suppress all output except errors and warnings (https://awscli.amazonaws.com/v2/documentation/api/latest/reference/s3/cp.html?highlight=quiet)

About your proposals, I like this one --> Add a new verbosity (e.g. SHELL_VERBOSITY=-2) that disables error output

But, I wouldn't do this:

Optionally, we might add some code in the bin/console recipe to default to -2 when using --quiet to remove this extra barrier for upgraded applications.

Having --quiet silently associated to a new verbosity different than quiet might look convenient but it might bite us in the future.

About possible naming of this new constant, AI suggested silent because "quiet" means that makes little or no noise and "silent" means that all noise is suppressed:

About possible naming of this new constant, AI suggested silent because "quiet" means that makes little or no noise and "silent" means that all noise is suppressed:

This makes sense to me. +1 for --silent from me.

Funnily enough, I encountered this issue yesterday. I was annoyed by the fact --quiet still displays the error. I wanted to remove all output and just rely on $?/$status to understand if the command failed.

Having a new --silent flag sounds like a good compromise. I'm 👍

silent 🤫 is better wording 👌🏻

Thanks for the feedback everyone!

PR is now updated with a new --silent option. Decided to not add a shortcut (-s), as this can be quite conflicting with shortcut options defined by commands (and Symfony itself already uses it in the CompleteCommand). Besides, when reading this discussion, the new mode is a special programmatic case anyways and we recommend people running commands manually to keep using -q instead.

Low deps failure is unrelated, flipped test failure will be fixed by #58320 if I understand correctly.

@kbond adding a new global option is always a BC break, as it'll break commands that defined this option before (you'll get an exception when registering that command).
I consider it a "minor backwards compatibility" that we allow in minor versions, but I thought it was good to flag it as BC break in the UPGRADE guide for better visibility.

Thank you @wouterj.