Module cli_version 

Detect the installed claude CLI version and (when reachable) the latest published version on npm.

The local probe shells out via forge_sdk::transport::process::query_cli_version and normalises the result to a bare MAJOR.MINOR.PATCH token - claude --version prints lines like 2.1.116 (anthropic) or claude 2.1.116 depending on the build channel, so callers want the version token alone for display.

CliVersionInfo is the cross-crate shape returned by the workspace mediator; both installed and latest are Option because either probe can fail independently (claude not on PATH, no network, npm rate-limited). The renderer treats None as a - placeholder so the panel’s row count stays constant.

Re-exports

use std::time::Duration; 🔒

Structs

CliVersionInfo

Snapshot of the installed-vs-latest claude CLI versions. Either side can be None independently; the renderer falls back to a dim - placeholder for the missing side.

Enums

ProbeOutcome 🔒

Result of one npm-registry attempt. The type exists for the retry decision: Unreachable never touched the server (retry direct), Unusable reached it but the response was no good (retry won’t help).

Constants

NPM_LATEST_URL 🔒

npm registry endpoint for the published @anthropic-ai/claude-code package. The /latest dist-tag returns the same JSON shape as the full package document but trimmed to the current latest version, which is all we need.

PROBE_TIMEOUT 🔒

Per-probe timeout (each of the two probes runs in parallel via tokio::join! inside crate::env::cli_version::fetch_info). 5 s is generous for a local --version invocation and still well below the user-visible “app feels stuck” threshold for the startup latency hit.

Functions

attempt_latest 🔒

Run one GET against the npm registry with the given client, classifying the result so the caller can decide whether a direct retry is worthwhile.

build_probe_client 🔒

Build the reqwest client for a single npm probe. direct adds .no_proxy() so the request bypasses any ambient proxy env and connects straight to the registry; both variants keep with_extra_roots so a corporate / mitmproxy CA still validates.

extract_semver_token 🔒

Extract the first whitespace-separated token that starts with a digit - handles 2.1.116 (anthropic), claude 2.1.116, and bare 2.1.116 shapes the CLI / npm produce.

fetch_info

Run both probes in parallel and return the merged snapshot. Always succeeds - failures collapse to None on the affected field with a WARN log; the workspace caller never has to handle a Result.

is_strictly_newer 🔒

true when lhs parses as a strictly-greater semver triple than rhs. A parse failure on either side collapses to false so the update banner only fires when the comparison is unambiguous.

latest_version_or_warn 🔒

Normalise an npm /latest response body to a version token, WARN + None when it carries none - the same silent-collapse shape as the installed probe’s miss.

parse_latest_version 🔒

Pull the MAJOR.MINOR.PATCH token from an npm /latest dist-tag response body’s version field.

parse_semver_triple 🔒

Parse MAJOR.MINOR.PATCH (ignoring any -pre.1 / +build suffix after the patch number) into a (u32, u32, u32) tuple suitable for direct tuple comparison.

probe_installed 🔒

Spawn claude --version on a blocking thread (the underlying helper uses std::process::Command), bound the wait to PROBE_TIMEOUT, and normalise the output to a bare MAJOR.MINOR.PATCH token. WARN log + None on any failure.

probe_installed_within 🔒

probe_installed with the wait as a parameter, so a test driving a real child can widen the budget rather than race machine load.

probe_latest 🔒

GET npm’s /latest dist-tag and return the newest published @anthropic-ai/claude-code version as MAJOR.MINOR.PATCH.

reported_version_or_warn 🔒

Normalise a zero-exit claude --version stdout to a bare version token, WARN + None when it carries none - a silent None here is indistinguishable from a probe that never ran.