[mparrett]

Two small asks from the reviews of #1076 and #1077, one commit each.

#1077, docs. With LETGO_USE_TINYGO=1 and -w-module, lg builds the tinygo arm's GOFLAGS from the environment variable and sets it. An environment GOFLAGS takes precedence over the go env file, so flags saved with go env -w drop out of that build. docs/guide/custom-lg.md now says so and tells users to export them instead.

#1076, tests. test/ir_mul_zero_test.lg now also checks (* x 0) on a ^long-hinted param and (* x 0.0), both under *ir-compile*. A ^long hint does not make x an Int: a double passed to it still comes back a double, which is why #1076 dropped the fold instead of guarding it on the type. The expected values match what bytecode returns for the same calls.

Verification

• make test LG_TEST=ir_mul_zero passes.

🤖 Generated with Claude Code

[mparrett]

Self-check. No actionable findings.

The documentation accurately describes tinyGoBuildEnv: for a caller-module build it constructs GOFLAGS from the supplied environment and appends -mod=readonly, so flags saved only in the Go env file are superseded. The added strict IR tests cover both multiplication operand orders, a ^long parameter hint, and a floating zero, including NaN and signed zero. Comparing printed values correctly distinguishes -0.0 from 0.0 and avoids NaN equality pitfalls. No production behavior changes.

Validation at 31a141d: the ir_mul_zero suite passed 3 tests / 6 assertions; TestTinyGoBuildEnv passed all four subtests; docs_frontmatter_hook.py --check passed for the changed guide. Independently checked the expected output vectors through ordinary bytecode: ["0.0" "NaN" "-0.0" "0"] for a ^long-hinted parameter multiplied by integer zero, and ["0.0" "NaN" "-0.0" "0.0"] for multiplication by floating zero. Existing CI is green.

[nooga]

Approving. The GOFLAGS note and the ^long/float-zero IR tests match bytecode behaviour.