`qbs session` uses CPU

[arBmind]

Observation:

• QtCreator spawns multiple qbs session processes
• On Windows each qbs session process consumes a .1-.3 % of my CPU when idle. With 4 processes this doubles the idle usage of my laptop and is a huge waste of battery.

Goals:

• Do not drain my battery / consume idle power
• Reduce extra processes and communication overhead

Analysis:

• qbs session communicates over stdin and stdout with host process (QtCreator)
• uses a QTimer to check if data was sent every 10ms
• A qbs session process is spawned for each project configuration (Up to 4 per project - I guess debug/release and 2nd config)
• clangbackend is not measurable on idle.
  • uses seperate named pipe (non blocking socket)
  • stdio is forwarded to qtcreator (not used for communication)
  • custom QDataStream based encoding (only useful for Qt environments)
• clangd used by VsCode is not measurable on idle.
  • simple use of blocking stdin
  • mitigates debugger interrupts (see clang-tools-extra/clangd/JSONTransport.cpp)
  • seems to spawn threads for actual work

Experiments on qbs:

• Even waiting for 90ms effectively brick qbs.
• Using singleShot timer might help for sleeping. But does not stop idle drain.

Perfect solution:

• Write a dedicated qbsd
• use approach from clangd with blocking stdin reading
• spawn/communicate with thread per session
• hope: qbscore was doing something similar when it was part of qtcreator?
• possible challenges:
  • protocol needs to be extended by something like a sessionId, start and end of sessions.
  • qtcreator needs to hold a common instance and despawn qbsd when not needed

Hacking solution:

• Use worker thread in qbs to read stdin blockingly
• Abort thread by closing stdin
• Process received data with signals

Alternatives:

• Use CompletionIO for Windows
  • Would require adding completion IO support to Qt main loop (might not be actually hard, but requires very careful design)