dev-util/nvidia-cuda-{sdk,toolkit}: Version bump to 8.0.44 #2445
Conversation
| # dodoc doc/*txt | ||
| if use doc; then | ||
| dodoc doc/pdf/* | ||
| dohtml -r doc/html/* |
There was a problem hiding this comment.
dohtml is deprecated in EAPI=6. Use the HTML_DOCS variable + einstalldocs instead. Consider doing the same with DOCS too (i.e. DOCS=( doc/pdf/. ))
There was a problem hiding this comment.
Already played around with this. The result is that either the directory structure gets messed up (no html subdirs anymore) or no documentation gets installed at all. I somehow don't get it..
There was a problem hiding this comment.
have you just tried
if use doc; then
DOCS+=( doc/pdf/. )
HTML_DOCS+=( doc/html/. )
fi
einstalldocs
?
There was a problem hiding this comment.
Setting DOCS and calling default at the end of the phase does not work, because files are previously cleaned up. Calling dodoc instead of dohtml messes up the file structure. Thus any hint which won't needs to redo the complete phase is appreciated. Thus I just left dohtml, since its depreciated but still doing its job.
There was a problem hiding this comment.
I missed calling einstalldocs and as mentioned calling default was not helping. Thanks for the hint, fixed now.
There was a problem hiding this comment.
'messes up file structure' -- sounds like you were doing something wrong. HTML_DOCS doesn't do anything except for calling dodoc -r…
| if use profiler; then | ||
| # hack found in install-linux.pl | ||
| for j in nvvp nsight; do | ||
| cat > bin/${j} <<- EOF |
There was a problem hiding this comment.
Done within new commit.
| UBUNTU_MENUPROXY=0 LIBOVERLAY_SCROLLBAR=0 \ | ||
| ${ecudadir}/lib${j}/${j} -vm ${EPREFIX}/usr/bin/java | ||
| EOF | ||
| chmod a+x bin/${j} |
There was a problem hiding this comment.
missing || die. Consider using fperms from Gentoo
There was a problem hiding this comment.
Added the missing die.
| fi | ||
|
|
||
| for i in ${remove}; do | ||
| ebegin "Cleaning ${i}..." |
There was a problem hiding this comment.
Done within new commit.
| dodir ${cudadir} | ||
| mv * "${ED}"${cudadir} || die | ||
|
|
||
| cat > "${T}"/99cuda <<- EOF |
There was a problem hiding this comment.
Done within new commit.
| ewarn "or append --compiler-bindir= pointing to a gcc bindir like" | ||
| ewarn "--compiler-bindir=${EPREFIX}/usr/*pc-linux-gnu/gcc-bin/gcc${b}" | ||
| ewarn "to the nvcc compiler flags" | ||
| echo |
There was a problem hiding this comment.
use ewarn instead of echo in order not to mix different output streams
| b=${a[${#a[@]}-1]} | ||
|
|
||
| # if gcc and if not gcc-version is at least greatesst supported | ||
| if [[ $(tc-getCC) == *gcc* ]] && \ |
There was a problem hiding this comment.
use the new tc-is-gcc from toolchain-funcs here instead
There was a problem hiding this comment.
Somehow done, hopefully the right way. Since I only have gcc 4.9 installed I cannot test this, due to modifying the cuda_supported_gcc variable.
| doman doc/man/man*/* | ||
|
|
||
| use debugger || remove+=" bin/cuda-gdb extras/Debugger extras/cuda-gdb-${PV}.src.tar.gz" | ||
| ( use profiler || use eclipse ) || remove+=" libnsight" |
There was a problem hiding this comment.
use a normal if here please, these ( ) subshell thingies aren't nice to read
There was a problem hiding this comment.
Done within new commit.
monsieurp
added
version bump
assigned
| # dodoc doc/*txt | ||
| if use doc; then | ||
| dodoc doc/pdf/* | ||
| dohtml -r doc/html/* |
There was a problem hiding this comment.
have you just tried
if use doc; then
DOCS+=( doc/pdf/. )
HTML_DOCS+=( doc/html/. )
fi
einstalldocs
?
marbre
force-pushed
the
nvidia-cuda
branch
2 times, most recently
from
588f576
to
70e182e
Compare
| LICENSE="CUDPP" | ||
| SLOT="0" | ||
| KEYWORDS="~amd64 ~amd64-linux" | ||
| IUSE="debug +doc +examples opencl +cuda" |
There was a problem hiding this comment.
Done within new commit.
| media-libs/freeglut | ||
| examples? ( | ||
| media-libs/freeimage | ||
| media-libs/glew:* |
There was a problem hiding this comment.
How do examples support multiple versions of glew?
There was a problem hiding this comment.
I checked the compiled examples with ldd and greped for 'glew', which is not linked. Since glew headers are also shipped with GL, I assume that glew is linked statically. Thus a recompile when glew is updated shouldn't be needed. Please correct me if my assumptions are wrong.
There was a problem hiding this comment.
Static linking is almost always a very bad idea. Even if that happens, you really want to rebuild every time glew is upgraded, in case the old version was vulnerable or buggy.
There was a problem hiding this comment.
Okay, changed it to :=. Hopefully, I one day get the thing with the slotted dependencies, thanks four your patience.
|
|
||
| src_unpack() { | ||
| unpacker | ||
| unpacker run_files/cuda-samples*run |
There was a problem hiding this comment.
Why are you calling unpacker twice?
There was a problem hiding this comment.
Also, you should check the underlying compression. If it happens to be xz, then you need to dep on xz-utils.
There was a problem hiding this comment.
Because you need to unpack the archive which includes several further archives. Archives are compressed using gzip.
There was a problem hiding this comment.
Please add a comment about that because otherwise it looks like a mistake.
| # epatch "${FILESDIR}"/${P}-asneeded.patch | ||
|
|
||
| sed \ | ||
| -e 's:-O2::g' \ |
There was a problem hiding this comment.
-O3 is removed in the next line of code. Of cause one could make a single line of code of it, but I think this is not a must.
| find common/inc/GL -delete || die | ||
| find . -type f -name "*\.a" -delete || die | ||
|
|
||
| default |
There was a problem hiding this comment.
I assume the default call is not necessary, thus should I drop it?
There was a problem hiding this comment.
If it doesn't have a configure script, then remove the default ;-).
There was a problem hiding this comment.
After removing it, I was reminded why I added it ;) Replaced it by eapply_user as either this or default has to be called.
|
|
||
| if use doc; then | ||
| ebegin "Installing docs ..." | ||
| treecopy $(find -type f \( -name readme.txt -o -name "*.pdf" \)) "${ED}"/usr/share/doc/${PF}/ |
There was a problem hiding this comment.
Also, don't indent on ebegin. It's confusing.
There was a problem hiding this comment.
Also if it is followed by an eend? I actually think I was told so in gentoo-sci.
| eend | ||
|
|
||
| ebegin "Moving files..." | ||
| for f in $(find .); do |
| doins ${f} | ||
| fi | ||
| fi | ||
| done |
| ebegin "Moving files..." | ||
| for f in $(find .); do | ||
| local t="$(dirname ${f})" | ||
| if [[ ${t/obj\/} != ${t} || ${t##*.} == a ]]; then |
There was a problem hiding this comment.
Why are you doing some crazy substitution instead of == with patterns?
| if [[ ${t/obj\/} != ${t} || ${t##*.} == a ]]; then | ||
| continue | ||
| fi | ||
| if [[ ! -d ${f} ]]; then |
There was a problem hiding this comment.
Actually the better way.. Done.
|
|
||
| if use doc; then | ||
| ebegin "Installing docs ..." | ||
| treecopy $(find -type f \( -name readme.txt -o -name "*.pdf" \)) "${ED}"/usr/share/doc/${PF}/ |
There was a problem hiding this comment.
Preferably by using "${ED%/}"/bla instead of "${ED}"bla
|
@mgorny @SoapGentoo Actually, its getting a little hard to bump ebuilds via pull request if there are thus many changes requested. Taking a look into what is in the tree, the ebuilds are already improved, cleaned up e.g. Thus I really don't know if I can spend enough time to take care for all requested changes, as I am also taking care for a bunch of ebuilds in the science overlay and mainly all jupyter ebuilds in the tree. |
Gentoo-bug: 595640 Package-Manager: portage-2.2.28
Gentoo-bug: 595366 Package-Manager: portage-2.2.28
|
@mgorny @SoapGentoo Most issues are resolved, hopefully in an acceptable way. The issues persists for ALL other cuda ebuilds in the tree which I didn't touched. Didn't changed a few thinks in the 8.0.44 ebuilds, since I don't know if there originally was a reason to implement it that way and, probably due to a lack of time, @jlec didn't yet respond. |
|
😞 The QA check for this pull request has found the following issues: Issues inherited from Gentoo (may be modified by PR): |
|
@SoapGentoo How can we step forward to get this finally merged? |
Gentoo-bug: 595366 Package-Manager: portage-2.2.28 Closes: #2445 Signed-off-by: David Seifert <soap@gentoo.org>
|
@marbre I've cleaned up the ebuilds and squashed them into your commits. I'm still not absolutely happy with the ebuilds, but they're in a passable state now. Please have a look at how I changed them (especially using bash arrays where possible). Let's try and keep them clean, lean and QA compliant. |
@SoapGentoo
This closes Gentoo Bug 595640 and https://github.com/gentoo-science/sci/issues/688. I'll adopted my release candidate ebuild from the overlay, droped the fetch restriction and further modified some lines. The required drivers is actually a guess, since Nvidia ships 367.48 which is not yet released. Anyway, with 367.44 deviceQuery from the sdk works.
Further, gcc 5.4 might be supported as bosted at parallelforall, but the currently shipped installation docs do not explicitly mention this.