Maximilian Roos 0b27755726 fix(help): name the doc-entry-point command from clap, not an argv scan (#3762) ...

Three developer entry points — `--help-page`, `--help-description`, and
`--print-schema` — each carried its own copy of a scan that picked the
subcommand out of argv by rejecting entries that looked like the binary:

```rust
.filter(|a| *a != "--help-page" && !a.starts_with('-') && !a.ends_with("/wt"))
.find(|a| !a.contains("target/") && *a != "wt")
```

Neither `ends_with("/wt")` nor `contains("target/")` matches `wt.exe`
under a backslash path, so on every Windows install all three read the
binary's own path as the command. `--print-schema` exited 2; the other
two exited 0 with empty stdout and `Unknown command: C:\...\wt.exe` on
stderr. The scan also took the value of a value-carrying global as the
command, so `wt -C <path> list --print-schema` answered `No JSON schema
for '<path>'`.

clap had already parsed the same argv one call earlier.
`parse_early_globals` runs the real `Cli` definition with
`ignore_errors(true)` and reads `matches.subcommand()` to pick the
alias-splice context, and the comment above it already said that exists
"so the splice path in `augment_help` has no separate arg scanner". The
doc entry points just never asked it. It now carries the name in an
`EarlyGlobals` struct, and the three handlers take `Option<&str>`
instead of `&[String]`.

Error messages are unchanged: `src/cli/mod.rs` has
`#[command(external_subcommand)]` for user aliases, so a name the `Cli`
definition doesn't declare still arrives intact and can be named back in
`Unknown command: X`.

`--print-schema` now prints through `crate::output::print_json` rather
than `serde_json::to_string_pretty` plus std's `println!`. std's panics
with exit 101 when a consumer closes the pipe; anstream's drops the
`BrokenPipe`. This was the last JSON-to-stdout surface still panicking
after #3746 — the schema landed via #3747 while that PR was in flight,
and it lives outside `src/commands/`, where #3746's stdout guard doesn't
scan.

<details>
<summary>Why CI never caught this</summary>

`tests/integration_tests/help.rs` and
`tests/integration_tests/readme_sync.rs` are both
`#![cfg(not(windows))]`, and on Unix the test binary is always named
`wt` under `target/`, which is exactly the shape the scan was written
against. The new test drives all four cases through a symlink named
`wt.exe` — a symlink rather than a copy because the debug binary is
67MB. Verified to fail against the pre-change code.

</details>

`wt merge --help-page` and `wt merge --help-description` still panic on
a closed pipe. Those use std's `println!`/`print!` deliberately, to
preserve the ANSI codes anstream's would strip, so they need a different
mechanism than this one. Left for a follow-up.

> _This was written by Claude Code on behalf of max-sixty_

2026-08-06 23:13:17 -07:00

16 KiB

Raw Permalink Blame History

View Raw