Add suport to is_integer/3 BIF (and to OTP-29)

[bettio]

OTP-29 introduces is_integer/3 BIF, so in order to support it we have introduced it as well.

• bif.c implements the new BIF. It checks first small integers in order to provide best performance with them.

• opcodesswitch.h: has been updated to support the new opcode for calling BIFs with 3 arguments, such as is_integer/3.

• intn has been changed in order to provide a helper for signed big integers comparison (now we have intn_cmpu and intn_cmp).

• README and documentation has been updated with OTP-29 support.

Closes #2114

These changes are made under both the "Apache 2.0" and the "GNU Lesser General
Public License 2.1 or later" license terms (dual license).

SPDX-License-Identifier: Apache-2.0 OR LGPL-2.1-or-later

[petermm]

The -export for id/1, classify/1, etc. was outside the -ifdef(OTP29_OR_LATER) block, so on OTP < 29 those functions were exported but never defined.

[petermm]

looks good:

Key Finding: Argument Validation Order Bug

bif_erlang_is_integer_3() validates arg1 (the term) before arg2/arg3 (the bounds). In OTP-29, if Min or Max is not an integer, it must raise badarg — even when Term is not an integer. The current code returns false early, diverging from OTP semantics outside of guards.

For example, erlang:is_integer(<<"x">>, foo, 100) should raise badarg, but currently returns false.

Recommended Fix

In bif.c, reorder validation:

Check arg2/arg3 are integers first → raise badarg if not
Then check arg1 → return false if not an integer
Then proceed to comparisons

Also:

Remove UNUSED(fail_label) since it's actually used by RAISE_ERROR_BIF
Add tests for is_integer(non_integer, bad_bound, good_bound) to pin OTP-29 failure semantics

[bettio]

erlang:is_integer(<<"x">>, foo, 100)

Thanks, everything fixed.

[petermm]

https://ampcode.com/threads/T-019ca09c-de95-774e-b3bb-bd023fcf800f

PR #2150 Review: is_integer/3 BIF (OTP-29)

#	File	Line	Severity	Issue
1	libs/jit/src/opcodes.hrl	171	Medium	OP_BIF3 (185) added to opcodes.h but missing from opcodes.hrl. Comment says "keep this list in sync". JIT won't handle the new opcode.
2	src/libAtomVM/opcodesswitch.h	7733	Medium	Copy-paste error: trace message says "bif2/2" but should be "bif3/3" in the OP_BIF3 handler.
3	src/libAtomVM/bif.c	259	Medium	UNUSED(fail_label) declared but fail_label is used by RAISE_ERROR_BIF on line 274. Misleading and may cause warnings.
4	CHANGELOG.md	83	Low	Typo: "OPT-29" should be "OTP-29".

and the formatting of course - otherwise good!

[bettio]

https://ampcode.com/threads/T-019ca09c-de95-774e-b3bb-bd023fcf800f

PR #2150 Review: is_integer/3 BIF (OTP-29)

Fixed everything. Thanks for reporting this.

[petermm]

llm strongly insists intn_cmp() should be intn_cmpu() here - ?

[petermm]

redundant outer parentheses..

[petermm]

absolute nitpicks left - looking good:

PR #2150 Review (v3): is_integer/3 BIF (OTP-29)

Commit: 1200210e6

Summary

The maintainer addressed all previous medium/high-severity issues:

• ✅ opcodes.hrl now includes OP_BIF3 and updated OPCODE_MAX
• ✅ Trace message fixed to "bif3/6"
• ✅ UNUSED(fail_label) removed (it's used by RAISE_ERROR_BIF)
• ✅ UNUSED(ctx) correctly omitted (ctx is used inside RAISE_ERROR_BIF macro)
• ✅ Test exports moved inside -ifdef(OTP29_OR_LATER)
• ✅ JIT support added in jit.erl with OP_BIF3 handler
• ✅ CI matrix extended with OTP-29 JIT build
• ✅ test_errors/0 now covers non-integer Term + non-integer bounds → badarg
• ✅ "OPT-29" typo in CHANGELOG fixed to "OTP-29"

Remaining Issues

#	File	Line	Severity	Category	Issue
1	src/libAtomVM/intn.h	283	Low	Doc	Confirmed (oracle): Stale docstring: "Magnitudes are compared using intn_cmp()" should be "intn_cmpu()" — the unsigned function was renamed but the doc for intn_cmp (signed) still references the old name. Not a self-reference; the implementation (intn.c:468) calls intn_cmpu().
2	src/libAtomVM/bif.c	274	Low	Style	return FALSE_ATOM; uses 2-space indent (10 spaces) but surrounding code uses 4-space indent (12 spaces). See line 271 RAISE_ERROR_BIF for the expected indentation level.
3	src/libAtomVM/bif.c	270	Low	Style	Redundant outer parentheses: `if ((!term_is_any_integer(arg2)

Notes

• The opcode gap (183–184 skipped, OP_BIF3 = 185) is correct — opcodes must match OTP's BEAM format numbering.
• The fast paths (small-int, int64) correctly fall through to the else block when any argument is non-integer, where proper validation occurs.
• RAISE_ERROR_BIF implicitly uses ctx via macro expansion (lines 55–56), so the ctx parameter is genuinely needed despite appearing unused at the source level.
• intn_cmp signed comparison logic is correct: sign mismatch → immediate result; same sign → unsigned magnitude compare, negated if both negative.
• Test coverage is comprehensive: small ints, int64, bigints, mixed types, guard context, direct call context, swapped intervals, single-point intervals, non-integer terms, non-integer bounds, and combined non-integer term + bounds.