-
Notifications
You must be signed in to change notification settings - Fork 140
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Implemented igraph #1196
Implemented igraph #1196
Conversation
If it helps generalize the PR, I believe the same |
Thanks very much. To merge this PR, we'll need to fix F2C_EXTERNAL_ARITH_HEADER so it doesn't contain user-specific paths. Maybe this can be done by patching the package's build script to set the variable, rather than setting it in meta.yaml. We'll also need a test script to verify that the package works. Please do the following:
|
server/pypi/build-wheel.py
Outdated
@@ -617,7 +617,8 @@ def env_vars(self): | |||
}) | |||
|
|||
for var in self.meta["build"]["script_env"]: | |||
key, value = var.split("=") | |||
key, value = var.split("=",1) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This was failing because of finding
IGRAPH_CMAKE_EXTRA_ARGS=-DF2C_EXTERNAL_ARITH_HEADER=../../../../../../../android64_x86_64_arm64.h -DIEEE754_DOUBLE_ENDIANNESS_MATCHES=ON
Multiple "=", now will only take the first segment as key and the rest as value as is
tested in Python 3.11, and ABIs:
Tested on ubuntu with pixel 4 API 31 emulator x86_64 By tested i mean i have a script the uses all that packages, and is working |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
python-igraph 0.11.6 is out now, you could consider packaging that version instead.
@@ -0,0 +1,9 @@ | |||
#define IEEE_8087 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Providing a header manually is OK if there's no simpler solution. But it should at least support x86_64 as well, because a lot of developers use emulators on that platform. That could be done with a single header file containing #ifdefs.
Also, is it really correct to define IEEE_8087 on ARM64?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Oh, I see from the filename that it's already supposed to cover both architectures. But there should probably still be some #ifdefs to only define the relevant things on each one.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also, is it really correct to define IEEE_8087 on ARM64?
Yes, this is needed for f2c to work. Don't take the name too seriously.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I guess you took this file from https://igraph.org/c/html/latest/igraph-Installation.html, which says it's for both ARM64 and x86, but on macOS. And all the Google search results for "IEEE_8087" do mostly return Apple-related things. So I'd like to see some evidence that these settings are also correct for Android.
It may also be possible to run the arithchk program mentioned in that link on an actual Android device to get the correct values. We've dealt with a similar situation before – see packages/chaquopy-hdf5/README.md.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The proper way to generate these headers is to run the arithchk
program, if you can. The reason why that's not done automatically here (and we need pre-generated files) is that you are cross-compiling.
This f2c-based code is just a fallback. f2c hasn't had a release since 1995 as far as I know. In an ideal situation, one links to an external BLAS/LAPACK/ARPACK, but we still want to keep igraph easy to compile with minimal external dependencies (partly for use cases like this one), so we do keep the f2c-based code functional and treat it as a first-class configuration (regular tests, etc.) But we don't want to touch the innards of f2c and fix things like the name of this IEEE_8087
macro
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We actually do have a build of OpenBLAS, and it's used by several other recipes. But I guess that doesn't include ARPACK, in which case f2c would still be required.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hi,
Back in the day, i was unsure to use header from the website so what i did is to run arithchk in my Samsung and in my emulator, both resulted the same also we are using this on many builds with QA passed from some months now.
(At least 3-4 emulators and at least 5 different android has run this)
Still please do all the checks you consider necessary.
Co-authored-by: Szabolcs Horvát <[email protected]>
It need to use ENV
IGRAPH_CMAKE_EXTRA_ARGS='-DF2C_EXTERNAL_ARITH_HEADER=/home/cbernal/Downloads/android64_x86_64_arm64.h -DIEEE754_DOUBLE_ENDIANNESS_MATCHES=ON'
During the execution of build-wheel.py
And needs to use api-level 24
IGRAPH_CMAKE_EXTRA_ARGS='-DF2C_EXTERNAL_ARITH_HEADER=/home/cbernal/Downloads/android64_x86_64_arm64.h -DIEEE754_DOUBLE_ENDIANNESS_MATCHES=ON' ./build-wheel.py -v --api-level 24 --python 3.11 --abi arm64-v8a igraph