fleet: put the bootstrapped Node on PANE_PATH (#1256)

On a host with no system Node, tools/install.sh bootstraps one into
~/.mosaic/node/current/bin and records it in ~/.profile. The fleet unit runs
`env -i ... bash --noprofile --norc`, so ~/.profile is never read — that is
deliberate — and _build_runtime_bin_prefix did not name the directory itself.
Runtime binaries are `#!/usr/bin/env node`, so the pane resolved `mosaic` and
then died on `env: 'node': No such file or directory` after an install that
reported success. Measured on a greenfield VM.

The existing `npm config get prefix` branch cannot cover it: that reports a
package prefix (~/.npm-global), never a Node runtime directory.

The test case asserts the property rather than the string — it runs the pane for
real with a Node-shebang `mosaic` and requires the pane to have executed. A PATH
substring check would pass on a fix that put the directory in the wrong position.
Red without the launcher change, green with it, rest of the suite unaffected.

packages/mosaic/framework/tools/fleet/start-agent-session.sh

@@ -268,6 +268,15 @@ esac
 _build_runtime_bin_prefix() {
   local candidates=()
   if [ -n "$MOSAIC_RUNTIME_BIN" ]; then candidates+=("$MOSAIC_RUNTIME_BIN"); fi
+  # A host with no system Node gets one bootstrapped here by tools/install.sh, which
+  # records it in ~/.profile. The fleet unit runs `env -i ... bash --noprofile --norc`
+  # by design, so ~/.profile is never read and the directory has to be named here.
+  # The npm probe below cannot cover this: it reports a package prefix
+  # (~/.npm-global), never a Node runtime directory. It is first so the bootstrapped
+  # runtime wins on a host that has both — that is the one the installer verified.
+  # Runtime binaries are `#!/usr/bin/env node`, so without this the pane resolves the
+  # binary and then dies on `env: 'node': No such file or directory`.
+  candidates+=("$PANE_HOME/.mosaic/node/current/bin")
   if command -v npm >/dev/null 2>&1; then
     local npm_prefix
     npm_prefix=$(npm config get prefix 2>/dev/null) || true

packages/mosaic/framework/tools/fleet/test-start-agent-session.sh

@@ -348,6 +348,89 @@ for blocked in LD_PRELOAD= BASH_ENV= MOSAIC_UNTRUSTED_SENTINEL=; do
   echo "$pane_environment" | grep -qF "$blocked" && fail "runtime pane received $blocked"
 done
 
+# #1256. On a host with no system Node, tools/install.sh bootstraps one into
+# ~/.mosaic/node/ and writes that directory to ~/.profile. The fleet unit runs
+# `env -i ... bash --noprofile --norc`, so ~/.profile is never read — correctly, by
+# design — and _build_runtime_bin_prefix does not list the bootstrap directory. Its
+# `npm config get prefix` branch cannot cover the gap either: the installer points
+# npm's prefix at ~/.npm-global, so that branch contributes the npm-global directory
+# and never the Node one, however it resolves.
+#
+# The property under test is not "the string is in PATH". It is that the pane can
+# EXECUTE a Node-shebang runtime binary — which is what `mosaic` is
+# (`#!/usr/bin/env node`) and what actually failed: measured on a greenfield VM as
+# `env: 'node': No such file or directory` after a clean install that reported success.
+#
+# So this case runs the pane for real and requires it to have run. A PATH-substring
+# assertion would pass on a fix that put the directory in the wrong position, and it
+# would keep passing if the pane later stopped running for some unrelated reason.
+: > "$TMUX_CALLS"
+HOME_NODE="$ROOT/bootstrap-node/.config/mosaic"
+write_generated "$HOME_NODE" "coder-node"
+NODE_PANE_HOME="${HOME_NODE%/.config/mosaic}"
+NODE_BOOTSTRAP_BIN="$NODE_PANE_HOME/.mosaic/node/current/bin"
+mkdir -p "$NODE_BOOTSTRAP_BIN"
+
+# The bootstrapped runtime. It records that it ran, which is the evidence this case
+# turns on: no node reachable from the pane means no marker.
+cat > "$NODE_BOOTSTRAP_BIN/node" <<'SHIM'
+#!/usr/bin/env bash
+set -euo pipefail
+env -0 > "${MOSAIC_HOME:?}/fleet/pane-environment"
+SHIM
+chmod +x "$NODE_BOOTSTRAP_BIN/node"
+
+# write_generated plants its symlinks under the MOSAIC_HOME it is given; here the
+# pane's HOME is the trusted parent, so the pane's view of "installed" is this
+# directory instead. `pi` is what #1241 resolves against PANE_PATH; `mosaic` is what
+# the pane then executes, and it is a Node script — not a bash script that would run
+# anywhere and quietly hide the defect.
+mkdir -p "$NODE_PANE_HOME/.npm-global/bin"
+ln -sf "$FAKE_BIN/pi" "$NODE_PANE_HOME/.npm-global/bin/pi"
+printf '#!/usr/bin/env node\n' > "$NODE_PANE_HOME/.npm-global/bin/mosaic"
+chmod +x "$NODE_PANE_HOME/.npm-global/bin/mosaic"
+
+# The npm branch is modelled ALIVE and still cannot close the gap, which is the
+# stronger statement. An earlier draft of this case tried to model npm as absent —
+# true on a real bootstrap host, where npm lives only in the Node directory — and it
+# refused to run anywhere npm is in the system path, i.e. most machines. It was also
+# the weaker claim: it would have proven only that a dead branch supplies nothing.
+#
+# On a bootstrap host the installer sets npm's prefix to ~/.npm-global. So even with
+# `command -v npm` true and the branch executing, `npm config get prefix` yields the
+# npm-global directory and never the Node one. The gap does not depend on whether
+# that branch runs.
+NODE_LAUNCHER_BIN="$ROOT/bootstrap-node-launcher-bin"
+mkdir -p "$NODE_LAUNCHER_BIN"
+ln -sf "$FAKE_BIN/tmux" "$NODE_LAUNCHER_BIN/tmux"
+ln -sf "$FAKE_BIN/npm" "$NODE_LAUNCHER_BIN/npm"
+
+/usr/bin/env -i \
+  "HOME=$NODE_PANE_HOME" \
+  "PATH=$NODE_LAUNCHER_BIN:/usr/bin:/bin" \
+  "MOSAIC_HOME=$HOME_NODE" \
+  "MOSAIC_TEST_TMUX_CALLS=$TMUX_CALLS" \
+  "MOSAIC_TEST_HOME=$NODE_PANE_HOME" \
+  "MOSAIC_TEST_NPM_PREFIX=$NODE_PANE_HOME/.npm-global" \
+  MOSAIC_TEST_FLEET_OWNER=123e4567-e89b-12d3-a456-426614174000 \
+  MOSAIC_TEST_EXECUTE_PANE=1 \
+  "MOSAIC_TEST_PANE_PID=$$" \
+  "$START" coder-node
+
+[ -f "$HOME_NODE/fleet/pane-environment" ] || \
+  fail "pane could not execute a Node-shebang runtime: $NODE_BOOTSTRAP_BIN is absent from PANE_PATH (#1256)"
+node_pane_environment=$(tr '\0' '\n' < "$HOME_NODE/fleet/pane-environment")
+# Colon-pad and match a whole element. A regex with `(^|:)` after `.*` looks like it
+# does this and does not: an anchor cannot match mid-pattern, so it silently requires
+# a leading colon and rejects the directory in FIRST position — which is exactly where
+# the fix puts it. That produced a failure reading "pane ran but PANE_PATH does not
+# carry <dir>" against a PATH whose first element was that dir.
+node_pane_path=":$(printf '%s\n' "$node_pane_environment" | sed -n 's/^PATH=//p' | head -1):"
+case "$node_pane_path" in
+  *":$NODE_BOOTSTRAP_BIN:"*) ;;
+  *) fail "pane ran but PANE_PATH does not carry $NODE_BOOTSTRAP_BIN (PATH=$node_pane_path)" ;;
+esac
+
 write_interaction_generated() {
   local home="$1"
   local agent="$2"