QUERY · ISSUE

MICROPY_PORT_DEINIT_FUNC called after gc_sweep_all

openby amirgonopened 2020-01-03updated 2020-01-07

I'm using MICROPY_PORT_DEINIT_FUNC to release hardware resources upon soft reset.
Unfortunately, it's called only after gc ram is deallocated by gc_sweep_all.
My port uses gc allocated ram, and when MICROPY_PORT_DEINIT_FUNC is called it's too late, gc ram was already deallocated.

https://github.com/micropython/micropython/blob/10709846f38f8f6519dee27694ce583926a00cb9/ports/esp32/main.c#L145-L153

Is there any particular reason gc_sweep_all is called before mp_deinit?
Is it reasonable to move gc_sweep_all after mp_deinit?

CANDIDATE · PULL REQUEST

py: Fix incorrect test for MICROPY_PORT_DEINIT_FUNC.

closedby stinosopened 2018-09-10updated 2018-09-24

Fix for #4112

Assessment

S Sonnet · high

Both items involve MICROPY_PORT_DEINIT_FUNC, but they address distinct bugs. PR #4114 fixes a preprocessor typo (MICROPY_PORT_INIT_FUNC used instead of MICROPY_PORT_DEINIT_FUNC) that prevented the deinit function from being called at all. Issue #5487 assumes the deinit function is called correctly but complains about call ordering: gc_sweep_all runs before mp_deinit, so gc-allocated resources are already freed when the port deinit hook fires. The PR does not fix the ordering problem described in the issue.

Suggested action

Leave issue #5487 open. It describes a separate ordering problem not addressed by PR #4114. The maintainer should consider whether gc_sweep_all should be deferred until after mp_deinit (and MICROPY_PORT_DEINIT_FUNC) completes. Link #4114 as prior context since it fixed the prerequisite that the function is actually called.

MICROPY_PORT_DEINIT_FUNC called after gc_sweep_all

py: Fix incorrect test for MICROPY_PORT_DEINIT_FUNC.

Keyboard