Fatalization of goto LABEL construct

[jkeenan]

This pull request supersedes #23677. (I am filing a new pull request in part because in my own GH repo I changed the name of the branch I wish to merge into blead; network search suggests that filing a new p.r. is the easiest way to proceed.) This commit message incorporates aspects of the earlier p.r.'s commit message.

I've been working on this project over the past month-and-a-half. I am now ready to propose a plan to implement the fatalization of the behavior currently guarded by warnings category deprecated::goto_construct. This pull request would resolve #23618.

This project is, in effect, a feature removal process, so we have to move carefully. If we remove a feature from core, we also have to remove code testing or otherwise relying on that feature. We have to remove documentation about that feature except insofar as we discuss the removal or suggest other ways to approach the business problem the feature was previously intended to solve. We then have to assess how the feature removal would affect Perl code out in the wild -- which may, in turn, require developing new programs (outside of the core distribution) to facilitate that assessment. I have done all that, so this pull request is being submitted in non-draft status with the hope that we can get it into blead well before the Feb 20 2026 (v5.43-8) "contentious changes" deadline for our next production release in May 2026.

On September 1 we completed one stage in this project. In #23659 (c00b151), we moved many of the code blocks in t/op/goto.t which we anticipate will not be affected by this fatalization to new file t/op/goto-sub.t.

The remaining stages then became:

• Preliminary code changes in the Perl guts, including removal of the existing deprecation warning and its replacement with a new exception. To support the new exception we'll probably have to update documentation in pod/perldeprecation.pod and perldoc -f goto. We should get make test_porting to PASS. We should identify tests in the test suite other than t/op/goto*.t which need to test for an exception rather than a warning. All of this should, IMO, be done before we tackle t/op/goto.t head-on. With the exception of needing more detail in the entry for the fatalization in pod/perldiag.pod, all of this has been done in this pull request.

• Getting the main test files to PASS. That file is t/op/goto.t, which now passes largely because I have simply stripped out all the code that pertained to the now fatal feature. The extent to which the code remaining in that file exercises ("covers") the remaining goto features needs investigation.

• Identifying the scope of Blead Breaks CPAN (BBC) problems. Since CPANtesters is set up on the premise that we're only testing CPAN distros against perl production/maintenance releases or commits to blead, we'll have to figure out a way to test against a branch without polluting the CPANtesters matrix. I have done that by revising one of my previous CPAN distributions and releasing it as Test-Against-Commit. I'll include details of my research using that library in a separate post.

Please review. Thank you very much.

---

• This set of changes requires a perldelta entry, and I need help writing it.

[tonycoz]

It would be nice if we could keep the more detailed messages here, though that would take some figuring out.

[jkeenan]

In blead, those other exception messages are defined in pp_ctl.c in the S_check_op_type function:

3229 S_check_op_type(pTHX_ OP * const o)
3230 {
3231     /* Eventually we may want to stack the needed arguments
3232      * for each op.  For now, we punt on the hard ones. */
3233     /* XXX This comment seems to me like wishful thinking.  --sprout */
3234     if (o == UNENTERABLE)
3235         croak(
3236                   "Can't \"goto\" into a binary or list expression");
3237     if (o->op_type == OP_ENTERITER)
3238         croak(
3239                   "Can't \"goto\" into the middle of a foreach loop");
3240     if (o->op_type == OP_ENTERGIVEN)
3241         croak(
3242                   "Can't \"goto\" into a \"given\" block");
3243 }
3244 

Note the inline comments and misgivings in the above -- all from b537774 by Father C back in 2017.

The warning about impending fatalization occurs later on in the same file, in PP(pp_goto):

3247 PP(pp_goto)
3248 {
...
3649         if (*enterops && enterops[1]) {
3650             I32 i = enterops[1] != UNENTERABLE
3651                  && enterops[1]->op_type == OP_ENTER && in_block
3652                     ? 2
3653                     : 1;
3654             if (enterops[i])
3655                 deprecate_fatal_in(WARN_DEPRECATED__GOTO_CONSTRUCT,
3656                         "5.42",
3657                         "Use of \"goto\" to jump into a construct");
3658         }
...
3705 }

In blead, we see these messages here:

$ ack 'into the middle of a foreach loop|into a binary or list expression|into a.+?given.+?block' *.c t/
pp_ctl.c
3236:                  "Can't \"goto\" into a binary or list expression");
3239:                  "Can't \"goto\" into the middle of a foreach loop");
3242:                  "Can't \"goto\" into a \"given\" block");

t/lib/croak/pp_ctl
7:Can't "goto" into the middle of a foreach loop at - line 3.
14:Can't "goto" into a "given" block at - line 3.
20:Can't "goto" into a "given" block at - line 2.
29:Can't "goto" into a binary or list expression at - line 2.
30:Can't "goto" into a binary or list expression at - line 3.
31:Can't "goto" into a binary or list expression at - line 4.
32:Can't "goto" into a binary or list expression at - line 5.

t/op/goto.t
286:    like($@, qr/Can't "goto" into the middle of a foreach loop/,

t/op/eval.t
228:	like($@, qr/Can't "goto" into the middle of a foreach loop/,

Now I acknowledge that I don't understand why, in blead, the deprecation warning is sometimes encountered first, whereas other times 1 of the 3 croak messages is encountered first.

In the branch, we see them here:

$ ack 'into the middle of a foreach loop|into a binary or list expression|into a.+?given.+?block' *.c t/
pp_ctl.c
3236:                  "Can't \"goto\" into a binary or list expression");
3239:                  "Can't \"goto\" into the middle of a foreach loop");
3242:                  "Can't \"goto\" into a \"given\" block");

t/lib/croak/pp_ctl
26:Can't "goto" into a binary or list expression at - line 2.
27:Can't "goto" into a binary or list expression at - line 3.
28:Can't "goto" into a binary or list expression at - line 4.
29:Can't "goto" into a binary or list expression at - line 5.

t/op/eval.t
228:	like($@, qr/Can't "goto" into the middle of a foreach loop/,

The 3 messages are still defined. 2 of the 3 are still exercised in either t/lib/croak.t or t/op/eval.t. The message about a given block is no longer exercised. If the point of this fatalization is to have an exception throw whenever we jump to a LABEL that is located inside a "construct" (a poorly defined term, IMO), then I don't see why we need the 3 more specific messages at all anymore. If you want to retain them and test for them, you should make an argument for that (though that would constitute mission creep, IMO).

But, as I acknowledged in earlier posts, I've never used goto myself, so I could be missing something.

[jkeenan]

Note: I welcome suggestions as to how we handle the entry in pod/perldiag.pod for this exception. So far I've just got a skeleton:

    Use of goto to jump into a construct is no longer permitted
        (F) More TO COME.

Further note: Once we implement this fatalization, do we still need these others?

    Can't "goto" into a binary or list expression
        (F) A "goto" statement was executed to jump into the middle of a
        binary or list expression. You can't get there from here. The reason
        for this restriction is that the interpreter would get confused as
        to how many arguments there are, resulting in stack corruption or
        crashes. This error occurs in cases such as these:

            goto F;
            print do { F: }; # Can't jump into the arguments to print

            goto G;
            $x + do { G: $y }; # How is + supposed to get its first operand?

    Can't "goto" into a "defer" block
        (F) A "goto" statement was executed to jump into the scope of a
        "defer" block. This is not permitted.

    Can't "goto" into a "given" block
        (F) A "goto" statement was executed to jump into the middle of a
        "given" block. You can't get there from here. See "goto" in
        perlfunc.

    Can't "goto" into the middle of a foreach loop
        (F) A "goto" statement was executed to jump into the middle of a
        foreach loop. You can't get there from here. See "goto" in perlfunc.

[tonycoz]

Further note: Once we implement this fatalization, do we still need these others?

As I said earlier, I think the more detailed messages are more useful - "construct" is vague, "into the middle of a foreach loop" is specific and gives the user a clue of exactly what they're doing that is forbidden.