doctor: surface recent Electron crashes with #583 pointer

Adds _doctor_check_recent_crashes, called from run_doctor before the
log-file section. When systemd-coredump shows ≥3 Electron crashes in
the last 7 days, surfaces a [WARN] with two workarounds (Settings
toggle, CLAUDE_DISABLE_GPU=1) and a link to the tracking issue.

Filters by the caller-supplied electron_path when entries match, falls
back to all-electron entries with a footnote when they don't (covers
AppImage's transient mount paths and other Electron apps installed
side-by-side).

Silent when coredumpctl isn't on PATH (non-systemd hosts), when there
are zero matches, or when the count is below threshold.

Co-Authored-By: Claude <claude@anthropic.com>
This commit is contained in:
aaddrick
2026-05-06 08:28:25 -04:00
parent 88676f44a6
commit 368f83490e
2 changed files with 156 additions and 0 deletions

View File

@@ -436,6 +436,66 @@ JSEOF
fi
}
# Surface a warning when systemd-coredump shows N+ recent Electron
# crashes. The most common cause on Linux is the GPU process FATAL
# exhaustion tracked in #583 — workaround for affected users is the
# upstream Settings → disable hardware acceleration toggle, or
# CLAUDE_DISABLE_GPU=1 in the environment for headless persistence.
#
# Arguments: $1 = electron path (e.g.,
# /usr/lib/claude-desktop/node_modules/electron/dist/electron)
# Used to filter results to claude-desktop's electron when possible;
# falls back to all-electron crashes when the path doesn't match
# (e.g., AppImage mount paths are transient).
_doctor_check_recent_crashes() {
local electron_path="${1:-}"
command -v coredumpctl &>/dev/null || return 0
# `coredumpctl list electron` filters by COMM=electron. If the
# exact electron_path matches any entry's EXE column, prefer that
# tighter count; otherwise fall back to all-electron entries.
local listing total_count path_count
listing=$(coredumpctl list electron \
--since='7 days ago' --no-pager 2>/dev/null) || return 0
[[ -n $listing ]] || return 0
# Drop the header line; count remaining entries.
total_count=$(awk 'NR>1 && NF>0' <<< "$listing" | wc -l)
((total_count == 0)) && return 0
if [[ -n $electron_path ]]; then
path_count=$(awk -v p="$electron_path" \
'NR>1 && index($0, p)' <<< "$listing" | wc -l)
else
path_count=0
fi
# Use the path-matched count when available; else the unfiltered
# count with a footnote so the user knows it may include other
# Electron apps (Slack, VSCode, etc.).
local count footnote=''
if ((path_count > 0)); then
count=$path_count
else
count=$total_count
footnote=' (some entries may be from other Electron apps)'
fi
if ((count >= 3)); then
_warn "Recent Electron crashes: $count in last 7 days$footnote"
_info \
'Most common cause: Chromium GPU process FATAL (#583).' \
'Try one of:'
_info ' Settings → toggle hardware acceleration off → restart'
_info ' or set CLAUDE_DISABLE_GPU=1 in the environment'
_info \
'Tracking:' \
'https://github.com/aaddrick/claude-desktop-debian/issues/583'
elif ((count > 0)); then
_info "Recent Electron crashes: $count in last 7 days$footnote"
fi
}
# Run all diagnostic checks and print results
# Arguments: $1 = electron path (optional, for package-specific checks)
run_doctor() {
@@ -923,6 +983,11 @@ print(len(servers))
fi
fi
# -- Recent crashes --
# Surfaces the GPU process FATAL pattern (#583) before users
# notice the in-app "Claude crashed repeatedly" prompt.
_doctor_check_recent_crashes "$electron_path"
# -- Log file --
local log_path
log_path="${XDG_CACHE_HOME:-$HOME/.cache}"

View File

@@ -229,3 +229,94 @@ _skip_gtk_query() {
[[ $output != *'[WARN]'* ]]
[[ $output != *'ibus-gtk3'* ]]
}
# =============================================================================
# _doctor_check_recent_crashes: GPU FATAL crash counter (#583)
# =============================================================================
# Install a coredumpctl shim. $1 is the coredumpctl-list-style
# multi-line output to emit (header + entry rows). The shim ignores
# its arguments — tests don't exercise the filter syntax.
_install_coredumpctl_shim() {
mkdir -p "$TEST_TMP/bin"
cat > "$TEST_TMP/bin/coredumpctl" <<SHIM
#!/usr/bin/env bash
cat <<'OUT'
$1
OUT
SHIM
chmod +x "$TEST_TMP/bin/coredumpctl"
export PATH="$TEST_TMP/bin:$PATH"
}
@test "_doctor_check_recent_crashes: no coredumpctl on PATH — silent" {
# Force coredumpctl off PATH so the helper short-circuits.
# Restore PATH before returning so teardown's rm works.
local saved_path="$PATH"
export PATH="/no-such-dir-for-test"
run _doctor_check_recent_crashes \
'/usr/lib/claude-desktop/node_modules/electron/dist/electron'
export PATH="$saved_path"
[[ $status -eq 0 ]]
[[ -z $output ]]
}
@test "_doctor_check_recent_crashes: zero crashes — silent" {
# Listing has the header line only, no entry rows.
_install_coredumpctl_shim 'TIME PID UID GID SIG COREFILE EXE SIZE'
run _doctor_check_recent_crashes \
'/usr/lib/claude-desktop/node_modules/electron/dist/electron'
[[ $status -eq 0 ]]
[[ -z $output ]]
}
@test "_doctor_check_recent_crashes: 1 crash — info line, no warn" {
_install_coredumpctl_shim 'TIME PID UID GID SIG COREFILE EXE SIZE
Wed 2026-05-06 08:00:21 EDT 130375 1000 1000 SIGTRAP present /usr/lib/claude-desktop/node_modules/electron/dist/electron 21.6M'
run _doctor_check_recent_crashes \
'/usr/lib/claude-desktop/node_modules/electron/dist/electron'
[[ $status -eq 0 ]]
[[ $output == *'Recent Electron crashes: 1'* ]]
[[ $output != *'[WARN]'* ]]
}
@test "_doctor_check_recent_crashes: 3+ crashes — warn + #583 pointer" {
_install_coredumpctl_shim 'TIME PID UID GID SIG COREFILE EXE SIZE
Wed 2026-05-06 08:00:21 EDT 130375 1000 1000 SIGTRAP present /usr/lib/claude-desktop/node_modules/electron/dist/electron 21.6M
Mon 2026-05-04 07:44:48 EDT 930532 1000 1000 SIGTRAP present /usr/lib/claude-desktop/node_modules/electron/dist/electron 22.8M
Sun 2026-05-03 14:34:10 EDT 567221 1000 1000 SIGTRAP present /usr/lib/claude-desktop/node_modules/electron/dist/electron 12.4M'
run _doctor_check_recent_crashes \
'/usr/lib/claude-desktop/node_modules/electron/dist/electron'
[[ $status -eq 0 ]]
[[ $output == *'[WARN]'* ]]
[[ $output == *'Recent Electron crashes: 3'* ]]
[[ $output == *'CLAUDE_DISABLE_GPU=1'* ]]
[[ $output == *'/issues/583'* ]]
}
@test "_doctor_check_recent_crashes: path mismatch falls back with footnote" {
# Three crashes from a DIFFERENT electron binary (e.g., Slack).
# Caller passes claude-desktop's electron path, which doesn't
# match — helper falls back to total count and adds the footnote
# so the user knows the count may be cross-app.
_install_coredumpctl_shim 'TIME PID UID GID SIG COREFILE EXE SIZE
Wed 2026-05-06 09:00:00 EDT 200001 1000 1000 SIGSEGV present /usr/lib/slack/electron 30M
Wed 2026-05-05 09:00:00 EDT 200002 1000 1000 SIGSEGV present /usr/lib/slack/electron 30M
Wed 2026-05-04 09:00:00 EDT 200003 1000 1000 SIGSEGV present /usr/lib/slack/electron 30M'
run _doctor_check_recent_crashes \
'/usr/lib/claude-desktop/node_modules/electron/dist/electron'
[[ $status -eq 0 ]]
[[ $output == *'[WARN]'* ]]
[[ $output == *'may be from other Electron apps'* ]]
}
@test "_doctor_check_recent_crashes: empty electron_path falls back" {
_install_coredumpctl_shim 'TIME PID UID GID SIG COREFILE EXE SIZE
Wed 2026-05-06 08:00:21 EDT 130375 1000 1000 SIGTRAP present /usr/lib/claude-desktop/node_modules/electron/dist/electron 21.6M'
# Caller didn't pass an electron_path — helper still counts and
# emits the info line based on the unfiltered total.
run _doctor_check_recent_crashes ''
[[ $status -eq 0 ]]
[[ $output == *'Recent Electron crashes: 1'* ]]
[[ $output == *'may be from other Electron apps'* ]]
}