don't gc_free() ringbuf if no heap; cleanup ringbuf #6214
Conversation
d98faf5 to
8dcbfb3
Compare
8dcbfb3 to
b9a4f7b
Compare
|
I think I am going to make ringbuf not have the idiosyncratic |
|
@jepler this is ready for review. It will go in the next 7.2.x release. |
|
New |
|
Isn't |
We were checking The fundamental bug fix here is just this commit: All the rest is fixing some other bugs, some latent, and then, being dissatisfied with Happy to discuss this tomorrow. |
|
Closing this to make a simpler change just for 7.2.x. Will do additional work for |
Fixes RP2040: board.UART() giving "Attempted heap allocation when VM not running." on reload #6213.
UART deinit might happen when there is no VM, during
reset_board_buses(), so check before trying to free a UART ringbuf that is on the heap. This UART deinit possibility was introduced in Allow multiple board buses #5422.There was a lot of inconsistency about how and whether a buffer could be supplied as the UART ringbuf. Make the code consistent across multiple ports.
ringbuf_init()should have been used several places where it was not.There is a fine distinction between the ringbuf buffer size in bytes, and its capacity in bytes, which is one less.ringbuf_init()used to take the capacity, but I made it take the actual buffer size, because that was less confusing when passing in an existing buffer, though inconsistent withringbuf_alloc(), which takescapacityand then allocates a buffer of one extra byte. Some discussion here about this point: Introduced ringbuf_init to perform static buffer initialisation #5813 (comment)(EDIT) Rewrote
ringbufto removecapacity == size -1limitation. Now the size is the capacity. Code is smaller and I think more readable, at the very small expense of keeping a count of used slots.I will see if this builds, and test it more tomorrow, to make sure I have the ringbuf fixes right. Draft for now.