kernel-2.eclass changes TOUT to be "$(T).tmp_gas_check" instead of "$(T)/.tmp_gas_check". This results in creating "/var/tmp/.../gentoo-sources-.../temp.tmp_gas_check" instead of "/var/tmp/.../gentoo-sources-.../temp/.tmp_gas_check". Reproducible: Always Steps to Reproduce: 1. emerge -av gentoo-sources # any sources 2. Ctrl-C after unpack (or wait until emerge completes) Actual Results: A file "/var/tmp/portage/sys-kernel/gentoo-sources-*/temp.tmp_gas_check" exists. Expected Results: A file "/var/tmp/portage/sys-kernel/gentoo-sources-*/temp.tmp_gas_check" must be "/var/tmp/portage/sys-kernel/gentoo-sources-*/temp/.tmp_gas_check" and thus deleted after emerge completes (with FEATURE="-keepwork -keeptemp").
Created attachment 246735 [details, diff] proposed fix
Created attachment 345398 [details] tout.patch Added proper quoting as well to make it more reliable. This patch will be part of a series of patches that will be sent to the mailing lists and committed all at once to the tree in the near future.
17 Apr 2013; Tom Wijsman <TomWij@gentoo.org> kernel-2.eclass: Added a warning after the variables that modifying other variables in the eclass is not supported, there is a chance that we will not fix resulting bugs. Fixes bug #421721. Clarify the default DESCRIPTION and make it not use versions, a directory with ebuilds that inherit this eclass may contain multiple versions and we also don't want to give the impression that a new directory needs to made if that's not the case. Fixes bug #445110. Clarified which patch depths are used in the normal output and error output when applying patches. Fixes bug #436402. Made sure .tmp_gas_check is created inside the temp folder, it accidentally created temp.tmp_gas_check instead. Fixes bug #336732. Make UNIPATCH_DOCS work again, install 0000_README document when using genpatches. Fixes bug #301478.
Your proposed fix would create it in the root directory if T were not set so that is unacceptable, I also have assumed the wrong T in my fix due to some misunderstand; so, we're back to square one. We need to implement the following functionality implemented in the Makefile: - If T is set, use $(T)/.tmp_gas_check - If T is not set, use .tmp_gas_check in the current directory. I have reverted above commit for now, until we have a better patch implemented.
(In reply to comment #4) > Your proposed fix would create it in the root directory if T were not set so > that is unacceptable, I also have assumed the wrong T in my fix due to some > misunderstand; so, we're back to square one. > > We need to implement the following functionality implemented in the Makefile: > > - If T is set, use $(T)/.tmp_gas_check > - If T is not set, use .tmp_gas_check in the current directory. Erm, how T can be unset? it *must* be set in ebuild env...
(In reply to comment #5) > Erm, how T can be unset? it *must* be set in ebuild env... Correct. I believe it's part of PMS. But it's not the same across pkg_* and src_* functions. Does it have to be for this change?
(In reply to comment #5) > Erm, how T can be unset? it *must* be set in ebuild env... Guys, you fell for the same mistake as I did... We're talking about "Makefile scope" here, therefore when the user runs `make` inside /usr/src/linux; so, we're not talking about "ebuild scope" or PMS here! I'll correct the summary to avoid further confusion. Note that the original code had single quotes and places an unprocessed $(T) in the Makefile, as it is supposed to be. We just need to adapt this to place a slash only if T is set, the other option is to make . a default value for T if that is possible in a Makefile.
Makefiles should use ${TMPDIR}, this one's de facto standard, and always exported per the PMS. Potentially with /tmp fallback.
cannot reproduce.
(In reply to Alice Ferrazzi from comment #9) > cannot reproduce. It's still reproducible: the same TOUT is used in arch/powerpc/Makefile (as of linux-4.10.3), the same sed lines are still in kernel-2.eclass. Note, that it only affects powerpc, so you need to be cross-compiling for powerpc or running under powerpc.
yes just noticed that is triggered with ppc from the code
Created attachment 471294 [details, diff] Patch handling unset T case as well Alice, please, find an updated patch which addresses Tom's concern about unset $(T).
Created attachment 737200 [details, diff] Slightly modified patch Andrey, can you please test this slightly modified version of your patch ?
> Andrey, can you please test this slightly modified version of your patch ? Mike, the patch looks good to me! Unfortunately, I can't easily test it anymore because I don't have PowerPC toolchain/sysroot anymore. But running sed command manually changes the Makefile as expected. While trying to test the patch, I found that it seems any leftovers of .tmp_gas_check got fully removed in kernel 4.20. So, there's no need (but no harm either) in running sed on kernels 4.20 and newer. Also, actual use of .tmp_gas_check got removed in 4.20 as well as 4.14.120, 4.19.15 stable kernels. However, 4.4 and 4.9 series still have it. Having said that, I'm not actually sure how Portage can now trigger `make` to touch .tmp_gas_check. The Makefile rule is triggered by "archprepare" target, so maybe it could get called on `emake headers_install` for linux-headers ebuild somehow. A couple more comments regarding the patch: > # only do this for kernel < 2.6.27 since this file does not exist in later > # kernels It seems the comment is stale or confusing. > + [[ -n ${KV_MINOR} ]] && ver_test ${KV_MAJOR}.${KV_MINOR}.${KV_PATCH} -lt 2.6.27 && \ > + ppc_makefile="${S}"/arch/ppc/Makefile Not a big deal, but it seems the indentation for ppc_makefile= line is off by 2 tabs.
Created attachment 737950 [details, diff] Latest patch after review Thanks for the visual review. I'v updated the patch but i'm not going to commit it until someone can test. That said, I'm going to close this bug until an interested party with the proper hardware wants to test. After that we can commit the patch attached
> Latest patch after review Sorry, looks like my previous comment confused it more than helped with the version checks. Let's probably do it this way: 1. If kernel version is < 2.6.27, the file to sed is "${S}"/arch/ppc/Makefile 2. If kernel version is >= 2.6.27 and < 4.20, the file to sed is "${S}"/arch/powerpc/Makefile 3. If kernel version is >= 4.20, no need to sed at all. > I'v updated the patch but i'm not going to commit it until someone can test. > > That said, I'm going to close this bug until an interested party with the proper hardware wants to test. > > After that we can commit the patch attached Fair enough.
