[architecture] Rework assertions to make them more useful #351
Conversation
salkinium
force-pushed
the
feature/new_assert
branch
10 times, most recently
from
75b2514 to
c188ded
Compare
|
This is ready for review. This is a breaking change in API, but not much change in behavior. The handlers and Tested on AVR, Cortex-M and Hosted-Darwin. |
| %% endif | ||
| uintptr_t context; | ||
| uint8_t behavior; | ||
| } _modm_assertion_info; |
There was a problem hiding this comment.
I need to call this from C code too, hence the packed struct once in C here and once in C++ as modm::AssertionInfo.
|
|
||
| #include "assert.h" | ||
| #include <modm/architecture/interface/register.hpp> | ||
| /// @endcond |
There was a problem hiding this comment.
On purpose, there is an issue somewhere where a @code hasn't been closed, but I couldn't find it in any of the header files. (at least none that were included here).
|
|
||
| %% if with_assert |
There was a problem hiding this comment.
Only include this if it is necessary, this is good for tiny targets.
| env.substitutions = { | ||
| "with_logger": True, | ||
| "with_assert": env.has_module(":architecture:assert") | ||
| } |
There was a problem hiding this comment.
We really need to deduplicate at least all the nucleo boards. This is getting ridiculous.
|
|
||
| KEEP(*(.assertion)) | ||
|
|
||
| __assertion_table_end = . + SIZEOF(.data); | ||
| __assertion_table_end = .; |
There was a problem hiding this comment.
This bug took me a day to find. I hate AVRs so much.
salkinium
force-pushed
the
feature/new_assert
branch
2 times, most recently
from
55c1788 to
15fe379
Compare
salkinium
force-pushed
the
feature/new_assert
branch
from
15fe379 to
c35d52c
Compare
This adds a reworked (breaking) model of assertions, where
modm_assert()forces abandonment, andmodm_assert_continue()allows for handlers to ignore assertions.This makes it possible to distinguish between the two at the assertion site.
In addition the assertion can have a short string, that is choses completely by the user and behaves just like a normal string, not the abomination with
\0as string delimiter before (I was being a little too smart lol). And a long description which is optionally included in the binary or placed in an ELF comment section to be looked up by the short name.Fixes #348.
TODO:
modm::ErrorReportwith assertions.vs Abort in libc. Abandon, since abort has slightly different semantics.Place descriptions into ELF comment section and integrate into build system.Better with a generic solution for logging.