Review requested:

• @nodejs/tsc

@marco-ippolito can you clarify what would be included, do you mean the general format applied to all console output messages or the addition/removal of anthing written to the console?

@marco-ippolito can you clarify what would be included, do you mean the general format applied to all console output messages or the addition/removal of anthing written to the console?

I mean the general format applied to all console output messages, for example the padding of console.table

Note that console methods delegate to util.format/util.inspect and https://nodejs.org/dist/latest-v21.x/docs/api/util.html#utilinspectobject-options

The util.inspect() method returns a string representation of object that is intended for debugging. The output of util.inspect may change at any time and should not be depended upon programmatically.

@targos I see your point
I think snapshot testing is something widely used (and I immagine something we want to support in our assertion library) and we definitely can't break those test with a patch like happened with #50135.

Could we change documentation stating that also the changes in util.format/util.inspect should be marked as breaking?

As @targos outlined, changes to the output format of console are intentionally defined as not breaking. We should not rely upon that output in case we need a stable output. It should as such also not be used to serialize input for snapshots. We have to define a stable output for things like that.

when you say a stable output you mean a new API that does something quite similar to util.format/util.inspect but its stable?

when you say a stable output you mean a new API that does something quite similar to util.format/util.inspect but its stable?

That would be a possibility, yes.

We could for example have an option as well that would output json or something else that would be stable (even though I believe having a separate API that is meant to be stable is better).

We could for example have an option as well that would output json or something else that would be stable (even though I believe having a separate API that is meant to be stable is better).

I'm having trouble to immagine how that would work for console.methods like table. Maybe we can decouple console output from util.inspect and format? But then we have to maintain 2 apis that do the same things

@BridgeAR @targos I copied the inspect functions to be an internal/console, we have to maintain now 2 separate implementation, but if it affects console we can now consider it a breaking change.

It doesn't change my position. I disagree that console.log output should be generally stable between versions.

It doesn't change my position. I disagree that console.log output should be generally stable between versions.

can you please expand on why the console output should not be considered stable?
I think many of our users rely on it for snapshots, benchmark, immagine the impact of a PR like this #51629 being shipped in a patch

@marco-ippolito you can add the dont-land labels to the other PR regardless of the status of this PR.

If we don't provide semver stability of console.table, then we should warn the user to not use it, because it would be impossible to have tests for it. A couple of my modules broke because of an output change that was backported (alignment).

I'm fine with that. console is a debugging tool, and is documented as such.

Here's an example of why I think it isn't realistic or desirable to consider all changes to console output to be breaking: #52283.
Note that what that PR does is not mitigated by having two separate implementations of the util.inspect function.

We might find consensus on a less drastic version of the change:

• Remove the duplication of util.inspect code. I am honestly surprised that people find okay to copy-paste and start maintaining separately more than 2000 lines of code.
• Document more specific behaviors as stable. For example:
  • What happens when specific types of arguments (like string) are passed
  • Color handling
  • Table layout

What I'm mostly against here, is the idea of freezing the entire console behavior and considering small improvements in the debugging experience of complex objects as breaking.

Here's an example of why I think it isn't realistic or desirable to consider all changes to console output to be breaking: #52283.
Note that what that PR does is not mitigated by having two separate implementations of the util.inspect function.

We might find consensus on a less drastic version of the change:

• Remove the duplication of util.inspect code. I am honestly surprised that people find okay to copy-paste and start maintaining separately more than 2000 lines of code.
• Document more specific behaviors as stable. For example:
  • What happens when specific types of arguments (like string) are passed
  • Color handling
  • Table layout

What I'm mostly against here, is the idea of freezing the entire console behavior and considering small improvements in the debugging experience of complex objects as breaking.

I understand your concerns, I agree we should document changes that we consider breaking so we dont have to keep 2 implementations.
I also would make CITGM required for console changes.
What I consider breaking:

• Changes in the table layout, like spacing and alignment
• Change in the color output
• Changing function signature.

What I would not consider breaking:

• Nested object structure

I'll take some time to think more on how we can handle this

What I consider breaking:

I'm not sure I agree with all of these.

Changes in the table layout, like spacing and alignment

We should explicitly document these as a debugging tool and mention the format is not stable across Node.js versions.

I am OK with conceding on the output of some of the methods (like console.log) under some cases (like only one parameter that is a string so formatting happens elsewhere).

Change in the color output

Ditto, that should not be breaking (and I'm not sure why it would break end users anyway)

Changing function signature.

This is just like any other API in Node.js (if we remove/break a function it absolutely should be semver-major)

I've been reading up in github here to learn about what happened with node.js to make it so that ever since version 20.15, it seems, console.error comes out red and console.warn comes out yellow. I also found even more interesting behaviors today:

• If you give more than one arg, the colorization doesn't take place. E.g. node -e 'console.error("abc", 1)' won't make anything red.

I appreciate that the behavior is useful in typical situations, but I would like to share how it has a negative impact on any software that already tries to make good use of colors. If we emit strings that have any text colors (such as util.inspect output!) then I would have to do a good deal of bending over backwards to make sure to reach in there to change text color resets to the appropriate value. In this case if I know I am writing some colorized output of util.inspect into console.warn, I would need to do a string replacement of \e[39m with \e[33m to change text color reset escape codes into yellow text color escape codes. This would need to change to become \e31m in the case of console.error.

Since that's pretty awkward, the overall effect for me and possibly a lot of the dev community will likely be to avoid using the console APIs due to the added complexity.

Updated with specific behaviors that should be considered stable, rewording, bikeshedding welcome

I just noticed this change as we upgraded nodejs recently, and I wanted to provide some extra context as someone who works at a monitoring company. (I don't have a strong opinion, just sharing my observations).

Let's say I have a CI job which does console.error('FOO BAR') - either my own code, or third-party.

As the output is now colorized by nodejs, this will show up in raw logs as [31m�[1mFOO BAR�[22m�[39m.

Now, suppose I want to grep the raw logs of the CI jobs which were ingested raw without color stripping.

• Before node 20.15.0 / 22.2.0 I was able to search for FOO in my query. I could have created a log monitor to alert me on any FOO encountered in logs.
• Post node 20.15.0 / 22.2.0 I need to refine my query as [1mFOO. If I had monitor on FOO, it would stop working.

(It's an actual known issue with my current CI / monitoring setup. To be honest though, the log ingestion pipeline should be stripping the colors IMO, or tokenizing better).

changed my mind