fix(learn): gate lessons on the learner's code, not the hint comments

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.

[Chemaclass]

🤔 Background

Related #1258

Every lesson checks "did the learner use this API", but the template each lesson writes carries that name in its own TODO/Hint lines. All nine gates passed on their own untouched template — and a lesson could be completed with an unrelated passing assertion:

functiontest_multiple_assertions() {
# TODO: Check that message contains "bashunit"# Hint: assert_contains "substring" "$message"
assert_same 1 1
}

gate (assert_contains): 1 match — on the comment
run --fail-on-risky: exit 0
=> "✓ Excellent! Lesson 2 completed!"

without ever using assert_contains, assert_matches or assert_not_empty, which is the whole lesson. Pairs with #1257: that closed the "no work at all" path, this closes the "wrong work" path.

💡 Changes

• One shared bashunit::learn::count_in_code helper counting matches outside comment lines, used by all nine gates rather than nine copies
• Two tests: the gates that name an assertion are not satisfied by their own template, and the helper itself ignores a hint while counting the same call in code

On the guard's scope

It covers gates naming an assertion. The others look for function set_up() or function data_provider_, which the template declares in code as a skeleton on purpose — those cannot distinguish template from solution and are not meant to; --fail-on-risky is what makes an unfilled skeleton fail.

An earlier version asserted the invariant over every gate and passed only because a stray break stopped it after the first file, where it would have been false for five of them. The version here also asserts it compared something, so an empty set cannot fake a pass. Mutation-checked: reverting the helper to a plain grep fails both tests.