fix(snapshot): two false passes in the placeholder path, and cover its behaviour

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.

[Chemaclass]

🤔 Background

Two bugs, both making a snapshot assertion report success while comparing nothing — the exact failure a snapshot test exists to prevent.

🐛 1. The placeholder fallback matched anything

Without perl, matching fell back to grep. grep applies a pattern per line, and a pattern containing newlines is read as several alternative patterns — so a multi-line snapshot whose placeholder sat on its own line contributed a bare .*, matching any line of any input.

This snapshot matched the string NOPE:

Report
::ignore::
Done

The existing comment noted grep couldn't span lines. It didn't notice grep matched everything instead.

Now awk, with RS set to a byte the input cannot contain so the whole value is one record — verified at parity with perl across all twelve behaviours. If neither perl nor awk exists, the assertion fails and says so rather than guessing.

🐛 2. Absolute test paths silently auto-recorded

resolve_file prefixed ./ unconditionally, so an absolute test path became a path relative to the caller's cwd. The real snapshot was never read, a stray one was recorded elsewhere, and every snapshot assertion in the run passed.

This is what made me disbelieve the first bug for an hour — my own probes were running against auto-recorded snapshots instead of the files I'd written.

🧪 Fourteen tests, half of them negative

A placeholder must narrow what is compared, never turn the assertion into one that always passes.

Snapshot	Matches	Does not match
a::ignore::b	a123b, ab	XXX, aXXXc
::ignore::tail	anythingtail	anythingelse
cost $5.00 (x) ::ignore::	that exact prefix	cost 999 (x) Z
multi-line span	A\n…\nZ	NOPE

All pass under both perl and awk.

resolve_file gained an optional source override so a test can call it at any stack depth — BASH_SOURCE[2] is the test file only when reached through assert_match_snapshot.

📖 Docs

docs/snapshots.md now states what a placeholder does not relax, with a matching/non-matching table, the multi-line rule, why --snapshot-update skips placeholder snapshots (verified), and the perl/awk requirement.

✅ Verification

make sa · make lint · bash build.sh bin -v → ✅ Build verified ✅ · 1733 sequential / 1692 parallel-simple-strict.