As I understand, it will be a little hard to use both libjpeg and jpeg-turbo simultaneously with this implementation?
Also jpeg-turbo's run and build cases needs to be fixed: in former binaries renamed to PREFIX/bin/*-turbo to avoid conflict and in latter static library is not installed.

It looks fine, though .if ${PORT_OPTIONS:MPNGTEST} is kinda ugly. But please do exp-run before committing this. The libpng major version changes is always PITA and brings many ports breakages.

I think that all 3 @dir keywords can be removed.

Closed by commit rD45915 (authored by @ak).

So, what is the consensus about:

• gzip to compressed(term, not compress(1)) change
• clarification of PORTREVISION reset. I'm agree with wblock's suggestion.
• Removal 'repo churn' phrase.

Sync with github.com/freebsd/freebsd-ports-graphics repo:

• Fix mix of tab and spaces, make indentation consistent
• Cleanup of extra-devd patch
  • Use true/false instead od 0/1 as return value from the bool function
  • Combine two similar checks
  • Rearrange logic to remove goto and gratuitous continue
  • Add todo comment about usb_id
  • Misc: s/LogMessageVerb/LogMessage/, wrap comment, lowercase two error messages

Improve the comment for keyboard's check.

76 Why removing uhid?

Good catch! This is my local change sneaked in. My keyboard has multimedia keys which attach to uhid and xorg busies the device file. On the other hand, I've never seen mouse with uhid device. Perhaps there is no harm to keep this change.

I think someone ment doc and not doceng :-)

Doh. Fixed, sorry for the inconvenience.

28 the example is a gzip'ed tarball.

Yes, but I think that equate 'standard compressed tarball' to the tar.gz is too narrow.
Sure, it's still most popular format in the ports tree, but tar.bz2 and tar.xz represent a significant percentage of distfiles too. I think they also can be called an 'standard compressed tarball'.

Is there a reason doceng@ is added, instead of a more wider audience?
I see no reason for doceng@ to be involved in review of such a trivial change.

I apologize, someone recommended me to add doceng@ as reviewer for handbook.

That is somewhat wrong, I don't think we should hardcode such urls, it would be better to tell people to go to /bugzilla and where to click.

While it looks somewhat ugly, this url is stable and shouldn't change in observable future. So it's convenience versus aesthetics.

Closed by commit rD45838 (authored by @ak).

Closed by commit rP370983 (authored by @ak).

Okay. This is without regard to whitespace:

Thank you.

Please do not use "package(s)", it's unnecessarily complicated and a legacy of lazy programming. "Packages" means "zero or more" and is simpler.

Fixed.

Thanks for the suggestions, patch updated.

Please commit

Done.
Should I also implement it for @shared-mime-info and perhaps remove @fmutil ?

Closed by commit rP370490 (authored by @ak).

Interesting that this doesn't cause any issues with 'make checksum' or 'make fetch' currently.

There is a bug in do-fetch target, it ignores SIZE. I think it was discussed on #portmgr-lurker.

Closed by commit rP369334 (authored by @ak).

• Use INSTALL_TARGET=install-strip instead of STRIP_CMD

It seems that in the meantime @bapt has added a few more keywords :)
Shouldn't shared-mime-info.ucl be changed and check-stagedir.sh modified too?
Or it's a special case?

22 couldn't @rmtry deal with this?

Oh, it somehow reverted to original, it should be: @unexec rm -rf /var/lib/xdm/authdir/* 2>/dev/null || true
Because directory name can be authfiles.? and authfile name is A*.
I was not able to find a way to delete file/dir with unknown name without @unexec rm. Will be glad to hear any suggestions.

• Use options helpers (properly)
• Convert to @sample
• Strip lib(s)
• Use new @dir syntax
• Remove as many @unexec/@exec as possible

Options helper conversion is a little off. You want: XDMSHELL_CONFIGURE_ENABLE= xdmshell

Fixed, thanks.

Style-wise, I'd block OPTIONS_SUB underneath OPTIONS_DEFINE and block XDMSHELL_DESC with XDMSHELL_CONFIGURE_ENABLE.

Hmm, is it documented somewhere?

More modest version of patch. Don't remove asserts from bsd.port.mk/bsd.sanity.mk

No, I mean what I said. The ports framework will call 'make install' for whatever installer target is set for the port. If the port ignores DESTDIR then it will splatter files over PREFIX. *Then* the framework tries to package it and the user is alerted. It's too late for the files touched on their system though.

That's what I've tried to say. The port can touch host's fs before invocation of package manager. The only way to prevent this is to never build on the host's fs.

Hmm, I was under impression, that package is always created before installation.
But it seems that pkg register is used instead. IMO, it's bad optimization.

No, pkg register also fails with error 74.
Did you mean that bad port can install/remove files to host's fs before pkg invocation?
Yes, NO_STAGE can prevents this from happening in some cases. There is no guarantees that
private port has NO_STAGE at all, though.

Many ports will build and install just fine without STAGEDIR set. Do it from a real server (not poudriere) and you won't even know there is an issue until it's already splatted files all over /.

Hmm, I was under impression, that package is always created before installation.
But it seems that pkg register is used instead. IMO, it's bad optimization.

My reasoning was that corresponding bits were removed from bpm, so even if user will keep private port, it will fails anyway with easily recognized error.
Ok then, can I remove NO_STAGE condition from devel/ice ?

Commited to xorg-dev repo.

Sure, I'll wait.

I thought it was fixed for all keywords because @sample/@rmtry were already implemented. Good catch, thanks.

--disable-gallium-intel is still needed for old mesa.

It should be wrapped in .if !defined(WITH_NEW_XORG) then. So we wouldn't forget to remove it with old xorg.

In D708#23, @bapt wrote:

Nope but it was already ignored by pkg because @stopdaemon is wrong and dangerous

Can you please add a few words about it to CHANGES?

• The --disable-gallium-intel option doesn't exist anymore and --with-gallium-drivers doesn't contain intel driver by default, this 'if' block can be removed
• The conversion from rm to @comment worked for me after I've added @comment lib/libGL.la and @comment lib/libglapi.la. You may also want to move comments before @dirrm* for esthetic reasons.
• Unfortunately I can't test main change right now, I don't have freebsd on rpi atm.

Closed by commit rD45491 (authored by @ak).

• Fix title
• Use wording suggested by wblock@

Commited, thanks for review.

Updated patch, now with check for empty directories.

Should we not also check for empty directories

The hook does that. I've uploaded the correct patch.
Sorry about that.

Several appear every month. I can continue to remove them, but IMO svn hook will be a better solution.