Refactor xtask to bind and compile for every target. #15
Conversation
|
I think this should be ready for a review. I still don't know how to manage the different bindings per chip, if we use different headers. Or we generate everything in the same bindgen file, and we use cfg / features to specify which signature per chip of they differ. |
|
In esp-wifi we have bindings per chip and use a feature + cfg to select the right binding. Probably we can do the same here |
- Add support for esp32s2 - Move headers file back to esp-mbedtls-sys - Update bindgen - Update esp-hal, esp-wifi
|
Sorry, this PR got quite lengthy.
TODO
Every library compiles successfully, but on Xtensa targets, I'm getting errors when trying to run an example. |
|
Maybe we need to pass https://gcc.gnu.org/onlinedocs/gcc/Xtensa-Options.html#index-mlongcalls to the compiler on Xtensa |
|
I've done a minimal testing on I don't know how rigorous we want to be on testing. Edit: I just realized how large the binaries generated are, compared to origin, I think this should be looked at. It's over twice the size. |
|
Using clang, I was able to reduce binary sizes a bit, but not to the same level as upstream: This PR @90568c3: With CLANG: Origin: I can't seem to figure out what else to do to reduce the binary size. I think they should still be optimized and reduced. |
|
Besides the lib size looks quite good to me. There is a warning about
I have no idea what causes the increased size. I thought it's just he The final binaries are smaller now but that might be mainly because of optimizations in esp-wifi itself. There is not much differences in how Clang is called (comparing the previous shell file and the output of the CMake crate). (But there are some differences - maybe that is the clue) |
On my side, I would get a
I do think it's mainly because of the improvements in Since I've looked a bit at the binaries, I hope this gives us some clue:
nmOrigin: This PR: objdumpOrigin: This PR: |
I guess the Maybe it's worth to try changing the original compile script to pass the exact same arguments to CMake and see if it produces the same libraries. I guess it will. |
|
Using the original Bash build script ( I really feel it's something in the configuration and now cmake is called, but I don't know what. I've tried changing the @bugadani Maybe you can help with this. |
I can throw in random ideas but I'm very unlikely to be much help here. I can see that the latest rev uses gcc and its toolchain files, while the main branch uses clang's toolchain file so I assume it also uses clang, but that's about all the difference I see. Are the archives actually meaningfully different or does one compiler just throw in a bunch of random zero-sized stuff and different metadata? What you pasted here is essentially identical |
|
I've dumped the bytes of the generated Using the same cmake toolchain file origin: this PR: |
|
It feels like you're not passing some flags esp-idf passes, e.g. |
- Fixes an issue caused by `cmake-rs` auto-generating C flags which caused a larger binary size than expected. - Remove ESP32-S2 from justfile since it won't compile for async
AnthonyGrondin
force-pushed
the
feat/improve-building-process
branch
from
76d7219 to
e9cec52
Compare
|
The issue ended up being caused by I haven't successfully got clang working to xtensa targets, hence whey they are still so large, but the last commit is still a considerable improvement. |
This fixes an issue where docs would be generated on a single line instead of being properly normalized and formatted
|
Sorry, this is a kind of a big PR. I'm not sure what's left to be done. I was hoping to get the compilation with clang to produce the same binaries as origin, which are the smallest, too, but I don't think it really affects that much the total binary size. Especially with I think eventually the compilation could be done in CI, when it detects a changes in the header files, so that users don't have to push their own compiled libs, to mitigate Supply chain attacks. |
| @@ -1,5 +1,5 @@ | |||
| [target.xtensa-esp32-none-elf] | |||
| runner = "espflash flash --monitor" | |||
| runner = "espflash flash --monitor --baud 921600" | |||
There was a problem hiding this comment.
Would be good to not have a hardcoded baud-rate here (and in the other places)
I've refactored the xtask process to be able to compile mbedtls without needing a bash script, to try and improve user experience and compatibility. Defining a config per chip also allows a finer grain control over which peripherals we provide an alternative for.
Using different configs will generate different bindings, which might be an issue.
What's left to be done:- binding fails when the config file is namedmbedtls_config.hsaying it cannot findtime.h.- Adding binding and compilation target for Xtensa chips.Usage
.cargo/config.toml
To generate bindings:
To compile libs: