SIGABRT crash on Linux with librust_pty.so loaded — two crashes in 22 minutes under memory pressure #55459

Summary

opencode 1.19.0 crashes with SIGABRT (signal 6, SI_TKILL) on Linux. Two crashes occurred ~22 minutes apart. The first ran ~24 minutes during an active coding session; the second died almost immediately after startup. Both times librust_pty.so (bun-pty) was loaded in the address space. The crash also coincided with system-wide memory pressure (swap thrashing) from multiple concurrent opencode instances each spawning their own MCP server processes.

Environment

• opencode version: v1.19.0 (installed via mise, Bun-compiled single binary)
• OS: Arch Linux (Omarchy), kernel latest
• Terminal: Hyprland + xdg-terminal-exec (TUI mode)
• Install: /home/gd/.local/share/mise/installs/opencode/1.19.0/opencode
• Command line: opencode --auto
• MCP servers configured: playwright-mcp, firecrawl-mcp, postgres-mcp, billion-context (each spawned per opencode instance as child node processes)

Crash details

Crash 1 (PID 5103, 20:27:21 IST)

Signal: 6 (ABRT) si_code: SI_TKILL
Command Line: /home/gd/.local/share/mise/installs/opencode/latest/opencode --auto

Stack (unsymbolized — stripped Bun binary):

#0  abort (libc.so.6 + 0x256ea)
#1  opencode + 0x37ca5d0
#2  opencode + 0x44a5425
#3  opencode + 0x4dfadda
#4  opencode + 0x4f820fc
#5  opencode + 0x47894c6
#6  opencode + 0x4efcbb4
#7-#9  (unmapped JIT frames)
#10 opencode + 0x23a5ec4
#11 opencode + 0x4e85a99

Modules without build-id: /tmp/.bcdfeffff5f5f37c-00000001.so, /tmp/.9adf6bfbe3f1ffa7-00000000.so, librust_pty.so

Crash 2 (PID 46140, 20:49:04 IST)

Same signal, different lower frames (higher in the call stack — likely a different code path to the same abort):

#0  abort (libc.so.6 + 0x256ea)
#1  opencode + 0x37ca5d0
#2  opencode + 0x44a5425
#3  opencode + 0x4dfadda
#4  opencode + 0x4f820fc
#5  opencode + 0x47894c6
#6  opencode + 0x4839641
#7  opencode + 0x483693f
#8-#11  (unmapped JIT frames)
#12 opencode + 0x23a5ec4
#13 opencode + 0x52b6182
#14 opencode + 0x52b46c8
#15 opencode + 0x551aacf
#16 opencode + 0x3c5eebd

Common frames across both crashes: opencode + 0x37ca5d0, + 0x44a5425, + 0x4dfadda, + 0x4f820fc, + 0x47894c6, + 0x23a5ec4, + 0x4e85a99.

System context at time of crashes

• 11GB RAM total, ~8GB swap used during the crash window (swap 8.2GB/23GB)
• Multiple concurrent opencode instances running (at least 3 identified in logs)
• Each instance spawns its own MCP server child processes (playwright, firecrawl, postgres, billion-context) — approximately 5-6 node processes per instance at ~90-115MB RSS each
• No OOM kill was issued; the system froze under memory pressure (swap thrashing) before OOM could act
• After the crashes, swap dropped to 0B and memory recovered to 7GB free

Possible contributing factors

1. bun-pty / librust_pty.so: loaded in both crash address spaces. v1.19.0 release notes mention "Validate PTY connection origins and tickets before checking whether a PTY exists" — a recent PTY change.
2. Memory pressure from MCP process fan-out: Each opencode instance spawns independent MCP server processes. With 3+ concurrent instances, this creates 15+ node processes totaling several GB, driving the system into swap thrashing.
3. JIT-compiled code: Anonymous .so modules in /tmp/ (Bun JIT) without build-ids — frames in this region could not be symbolized.

Reproduction

Not reliably reproducible — occurred twice in 22 minutes under memory pressure with multiple concurrent instances. No specific user action immediately preceded either crash.

Request

1. Symbolize the frames above if possible (the binary is stripped; debug symbols unavailable via debuginfod).
2. Investigate whether the PTY validation change in v1.19.0 could abort under memory pressure or concurrent PTY access.
3. Consider bounding MCP server process fan-out when multiple opencode instances run concurrently on the same machine (shared MCP server pool or connection reuse).