term-readkey-pm build failure with Xcode11 #461

nieder · 2019-09-15T16:57:49Z

On 10.15b7 with Xcode GM, term-readkey-pm fails with this error:

cc  -I/Applications/Xcode-beta.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.15.sdk/System/Library/Perl/5.18/darwin-thread-multi-2level/CORE -bundle -undefined dynamic_lookup -fstack-protector  ReadKey.o  -o blib/arch/auto/Term/ReadKey/ReadKey.bundle  \
	      \
	  
chmod 755 blib/arch/auto/Term/ReadKey/ReadKey.bundle
"/usr/bin/arch" -x86_64 perl5.18 "-Iblib/arch" "-Iblib/lib" ReadKey_pm.PL ReadKey.pm 
Creating ReadKey.pm
Can't load 'blib/arch/auto/Term/ReadKey/ReadKey.bundle' for module Term::ReadKey: dlopen(blib/arch/auto/Term/ReadKey/ReadKey.bundle, 1): no suitable image found.  Did find:
	file system relative paths not allowed in hardened programs at /System/Library/Perl/5.18/darwin-thread-multi-2level/DynaLoader.pm line 194.
 at /System/Library/Perl/5.18/darwin-thread-multi-2level/DynaLoader.pm line 98.
	DynaLoader::croak('Can\'t load \'blib/arch/auto/Term/ReadKey/ReadKey.bundle\' fo...') called at /System/Library/Perl/5.18/darwin-thread-multi-2level/DynaLoader.pm line 194
	DynaLoader::bootstrap('Term::ReadKey') called at ReadKey_pm.PL line 515
Bootstrapping the XS for blockoptions: make: *** [ReadKey.pm] Error 255

Looks like more 'security' from Apple.
@cschaffner
@dmacks

The text was updated successfully, but these errors were encountered:

dmacks · 2019-09-16T03:38:51Z

Apple's new "hardened runtime entitlements" security apparently prevents a whole host of formerly allowed and commonly done and useful things that use "other than actual full paths to loadable objects". Apparently it also breaks @executable_path that for a while was promoted as allowing prefix-agnostic applications, DYLD env-var games (hello libtool/cmake!), and module-loading.

nieder · 2019-10-06T11:12:23Z

Just to add, extutils-constant-pm also fails with a hardened error on 10.15/xcode11, but just during tests, not pkg building.

cschaffner · 2019-10-06T14:57:04Z

@cschaffner
wrong person, I'm afraid...

nieder · 2019-10-13T12:17:38Z

For the record, I forced perl5184 on 10.15, symlinked perl5.18->perl5.18.4, and built term-readkey-pm5184 without any problems. Time to forgo system-perl on 10.15? It's probably going to be gone on 10.16 anyway.

dmacks · 2019-10-19T07:52:09Z

@nieder I started fink/fink#201 regarding replacing system-perl

TheSin- · 2019-12-09T14:35:04Z

I'm stuck at this bug so I'm going to try a few things, one is maybe during build it can set and alias for perl5.18.4 ? Anyhow I'll report back since this is holding up my work and I have no choice since I can only run 10.15 now.

TheSin- · 2019-12-09T15:26:11Z

alias failed, but maybe I'll spend more time understanding the actual issue, I assume it's in the perl binary it's self

@neider you forced perl5184 but why the symlink? Also any chance or adding perl5184-core to fink so we 10.15 ppl can build it too?

TheSin- · 2019-12-09T16:06:19Z

so really the issue is just using relative paths.

perl5.18 "-I/opt/sw/src/fink.build/term-readkey-pm5184-2.38-0.1/TermReadKey-2.38/blib/arch" "-I/opt/sw/src/fink.build/term-readkey-pm5184-2.38-0.1/TermReadKey-2.38/blib/lib" ReadKey.pm.PL ReadKey.pm
Creating ReadKey.pm
Bootstrapping the XS for blockoptions: 5
Done

changing the -Iblib/lib to -I/opt/sw/src/fink.build/term-readkey-pm5184-2.38-0.1/TermReadKey-2.38/blib/lib fixes it

TheSin- · 2019-12-09T16:15:15Z

Looks like 2 solutions that I can that are immediate.

We patch fink so that after Makefile.PL we run a perl -pi on Makefile that replaces ' blib/' with '$(pwd)/blib/'
we patch ExtUtils so that it uses absolutes

I think 2 is best but I'm not that familiar with ExtUtils, so that would take me a little time, hoping someone with more exp with ExtUtils can chime in here.

TheSin- · 2019-12-09T16:21:03Z

fixed upstream I'm going to locally update and test

Changes for version 7.37_04
Bug fixes:
Fix static linking on macOS

The 10.15 installed version is 6.66, like really??

I updated the fink version to 7.38 but I can't figure out how to force it to be used over the system version :\

nieder · 2019-12-09T16:53:24Z

@TheSin-: I made the symlink because I wasn't 100% sure how Fink would deal with two version of perl-5.18.4 around. I know a lot of the internal Fink code looks for the executable perl5.18, so I figured that might be needed (but is probably the wrong 'solution'). So since perl5184 creates %p/bin/perl5.18.4, I then symlinked %p/bin/perl5.18 to point to that (thus hiding /usr/bin/perl5.18).

To get perl5184-core on 10.15, I just edited dists/stable/main/finkinfo/languages/perl5184.info to have 10.15 in the Distributions: field.

TheSin- · 2019-12-09T16:56:56Z

Got it working, not pretty but works

in PkgVersion.pm

change

                                "$archflags $perlcmd Makefile.PL \%c\n".
                                "/usr/bin/make CC=gcc CXX=g++\n";

to

                                "$archflags $perlcmd Makefile.PL \%c\n".
                                "perl -pi -e 's, blib/, %b/blib/,g' Makefile\n".
                                "/usr/bin/make CC=gcc CXX=g++\n";

Obviously this type of fix could break other packages, it might be safer to do this inside the term-readkey-pm info file, but I'm going to test it in the PkgVersion.pm for a bit see how it goes.

This does NOT fix tests, this likely needs to be done in the test module as far as I can see.

TheSin- · 2019-12-09T17:17:41Z

To get perl5184-core on 10.15, I just edited dists/stable/main/finkinfo/languages/perl5184.info to have 10.15 in the Distributions: field.

Thank I got it working with my patch form above, I'm just fixing my debhelper now since my patches never accounted for multilevel base path ;)

htodd · 2019-12-13T18:58:06Z

This is working for me now.

nieder · 2019-12-14T07:21:32Z

Patching ExtUtils::Makemaker to use a full blib path would push a versioned BuildDepends on every perl package that uses ExtUtils::Makemaker. Then again, editing PkgVersion.pm would push a versioned Bdep on fink, so not much difference.

Doing perl-pie inside term-readkey-pm.info would mean breaking up %{default_script} into its individual commands to squeeze the perl-pie in between Makefile generation and running make. I think it's better to edit PkgVersion.pm.
To edit extutils::makemaker, MM_Any.pm:line1915 sets ./blib/arch.

$self->{INST_ARCHLIB} ||= $self->catdir($Curdir,"blib","arch");

No idea how difficult it'll be to change $Curdir to become a full path vs just .

TheSin- · 2019-12-14T13:47:03Z

Fink is essential which means you wouldn’t need a version build dep on it since it doesn’t change the dev content and it would auto install. But makemaker would need one

dmacks · 2020-02-04T16:18:52Z

I think patching MakeMaker is the cleaner solution...put the workaround in the place where the apparent bug (or at least "the call that is now broken by apple's changes") is. Does this affect other packages besides term-readkey-pmXXX (like "all other -pmXXX that have compiled C code)? If so, we could have fink insert an implicit BDep:eu-mm-pmXXX on all packages that are [Package:*-pmXXX && Type:makemaker], so packages themselves wouldn't have to do anything special.

TheSin- · 2020-02-04T16:37:03Z

it does affect others, but it's not all. I forget some of the others, I found it by patching term-readkey by hand and continue rather then patching fink.

nieder · 2020-03-28T00:58:16Z

See Perl-Toolchain-Gang/ExtUtils-MakeMaker#350

nieder · 2020-05-01T09:09:48Z

Aha!
This fixes extutils-makemaker so that term-readkey-pm5184 and extutils-constant-pm build on 10.15:

--- MM_Any.pm
+++ MM_Any.pm
 my($self) = shift;
 
+$self->{Curdir_Abs} = File::Spec->rel2abs ($Curdir);
-$self->{INST_ARCHLIB} ||= $self->catdir($Curdir,"blib","arch");
+$self->{INST_ARCHLIB} ||= $self->catdir($self->{Curdir_Abs},"blib","arch");
 $self->{INST_BIN}  ||= $self->catdir($Curdir,'blb','bin');

nieder · 2020-05-01T10:16:42Z

Ugh. Almost. EU::MM tests then compare abs path from patched MM_Any.pm to reference relative path.

nieder · 2020-05-01T12:07:27Z

EU-MM-7.44.zip

Can someone else on 10.15 check the above version of EU-MM ? I had to zip it up because Github doesn't support uploading .info or .patch to an ticket.

dmacks · 2020-05-01T16:09:58Z

I only have 10.13, and occasionally 10.14. Github supports .txt, so maybe rename to .info.txt and so on? That's 100% differently annoying than zipping it all:)

nieder · 2020-05-01T16:26:52Z

Also, current 10.15/pm5182 that use EU::MM often fail looking for EXTERN.h. I found this to be caused by applying the eu-mm-pm-sdk.patch file indiscriminately to all EU:MM variants on 10.14+. It should only be applied to the systemperl variant (pm5184 on 10.14.5 and 10.15).

If we expand the conditionalizing to test for $darwin_vers AND %type_num[perl], then the four pm5182 perlmods I tested that previously failed on 10.15/pm5182 now build and test OK.

htodd · 2020-05-01T21:37:01Z

extutils-makemaker-pm5184 works for me on 10.15.4. No errors with -Kk

Fix how EU::MM calculates INST_ARCHLIB to deal with hardened system-perl on 10.15 Patch using rel2abs to force an absolute path for the variable. Fixes fink#461

@TheSin-

The TermReadKey perl package fails to build due to a restriction apple added on dealing with .bundle files by their relative path name. This was diagnosed in 2019 by fink and I am adapting a workaround that @TheSin- proposed then: fink/fink-distributions#461 (comment) Also: - bump the sub-resources to their most recent versions while here - build failed until I added Devel::CheckLib too

@TheSin-

The TermReadKey perl package fails to build due to a restriction apple added on dealing with .bundle files by their relative path name. This was diagnosed in 2019 by fink and I am adapting a workaround that @TheSin- proposed then: fink/fink-distributions#461 (comment) Also: - bump the sub-resources to their most recent versions while here - build failed until I added Devel::CheckLib too Closes #66699. Signed-off-by: chenrui <[email protected]> Signed-off-by: BrewTestBot <[email protected]>

nieder added the bug label Sep 15, 2019

leejo mentioned this issue Mar 3, 2020

Test fails using built-in Perl 5.18 on macOS Catalina leejo/string-crc32#4

Open

nieder added a commit to nieder/fink-distributions that referenced this issue May 2, 2020

extutils-makemaker-pm 7.44

acd5a7d

Fix how EU::MM calculates INST_ARCHLIB to deal with hardened system-perl on 10.15 Patch using rel2abs to force an absolute path for the variable. Fixes fink#461

nieder mentioned this issue May 2, 2020

Fix extutils-makemaker-pm to deal with hardened system-perl on 10.15 #617

Merged

nieder closed this as completed in #617 Oct 12, 2020

mitchblank mentioned this issue Dec 11, 2020

innotop: fix build on 10.15+ Homebrew/homebrew-core#66699

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

term-readkey-pm build failure with Xcode11 #461

term-readkey-pm build failure with Xcode11 #461

nieder commented Sep 15, 2019

dmacks commented Sep 16, 2019

nieder commented Oct 6, 2019

cschaffner commented Oct 6, 2019

nieder commented Oct 13, 2019

dmacks commented Oct 19, 2019

TheSin- commented Dec 9, 2019

TheSin- commented Dec 9, 2019 •

edited

Loading

TheSin- commented Dec 9, 2019 •

edited

Loading

TheSin- commented Dec 9, 2019

TheSin- commented Dec 9, 2019 •

edited

Loading

nieder commented Dec 9, 2019

TheSin- commented Dec 9, 2019 •

edited

Loading

TheSin- commented Dec 9, 2019

htodd commented Dec 13, 2019

nieder commented Dec 14, 2019

TheSin- commented Dec 14, 2019

dmacks commented Feb 4, 2020

TheSin- commented Feb 4, 2020

nieder commented Mar 28, 2020

nieder commented May 1, 2020

nieder commented May 1, 2020

nieder commented May 1, 2020

dmacks commented May 1, 2020

nieder commented May 1, 2020

htodd commented May 1, 2020

term-readkey-pm build failure with Xcode11 #461

term-readkey-pm build failure with Xcode11 #461

Comments

nieder commented Sep 15, 2019

dmacks commented Sep 16, 2019

nieder commented Oct 6, 2019

cschaffner commented Oct 6, 2019

nieder commented Oct 13, 2019

dmacks commented Oct 19, 2019

TheSin- commented Dec 9, 2019

TheSin- commented Dec 9, 2019 • edited Loading

TheSin- commented Dec 9, 2019 • edited Loading

TheSin- commented Dec 9, 2019

TheSin- commented Dec 9, 2019 • edited Loading

nieder commented Dec 9, 2019

TheSin- commented Dec 9, 2019 • edited Loading

TheSin- commented Dec 9, 2019

htodd commented Dec 13, 2019

nieder commented Dec 14, 2019

TheSin- commented Dec 14, 2019

dmacks commented Feb 4, 2020

TheSin- commented Feb 4, 2020

nieder commented Mar 28, 2020

nieder commented May 1, 2020

nieder commented May 1, 2020

nieder commented May 1, 2020

dmacks commented May 1, 2020

nieder commented May 1, 2020

htodd commented May 1, 2020

TheSin- commented Dec 9, 2019 •

edited

Loading

TheSin- commented Dec 9, 2019 •

edited

Loading

TheSin- commented Dec 9, 2019 •

edited

Loading

TheSin- commented Dec 9, 2019 •

edited

Loading