Chris Billington wrote:
Stuart Henderson wrote:
On 2026/09/10 16:42, Chris Billington wrote:
+EXTRACT_SUFX = .tar.gz
not needed, this is the default
+SHA256 (3b12bc579b99932e3fe307ea1e44b8c1c6d1d5c9.tar.gz) =
8F0xFe9bimf51zwRpZdy3EKaUdcFCZ1vuUFOq5VLb1g=
not a valid filename for distfiles
the patch to PLIST doesn't fully apply:
Hunk #22 failed at 1981.
1 out of 124 hunks failed--saving rejects to pkg/PLIST.rej
On 2026/09/10 03:07, Anthony J. Bentley wrote:
Chris Billington writes:
- upstream uses git submodules for pugixml, SingleApplication, and the
'elements' library. I have used the same pinned revisions in
SITES.xxx/DISTFILES.xxx as the upstream branch.
DIST_TUPLE would be a better fit for this.
yes, that would fix the 3b12bc5... filename too. (btw you can use GH_*
for the main file and DIST_TUPLE for submodules; portroach is likely to
cope better with that)
- The 'elements' user-submitted library (CC-BY-SA 3.0) is pinned
against
a 9-month-old commit of the submodule repo.
Is it permissible to use a recent commit instead, since that would
still
make a reproducible build but would include more recent items?
If doing so provides value and doesn't introduce incompatibility, then
sure.
agreed (in many cases there is a reason why upstream pins a particular
version, in some other cases there isn't, so do check)
- portcheck notes 3 Makefile lines longer than 80 chars, but it's not
easy to break them without sacrificing readability. Is that acceptable?
Yes.
Attached revised diff (gzipped)
- use DIST_TUPLE (found it was not possible to mix DIST_TUPLE and GH_*)
- elements submodule now pinned to a more recent Aug 2026 commit: brings
in a lot of networking/dev board library elements
- diff applies OK to -current ports tree of 11 Sep, but I also attach
the port tgz (from the port directory) as a backup.
- Qt6: I tried building against Qt6/KF6, but the 0.100 branch is not yet
Qt6-clean. Specifically, cmake/fetch_kdeaddons.cmake is hardcoded to use
KF5. The upstream master branch is KF6/5 agnostic and it seems the next
0.200.x release will be Qt6-capable. Work is ongoing, Windows-first. So
we could maybe switch after next release.
- port-lib-depends-check marked GL and sqlite3 as 'Extra:' WANTLIBS for
the cmake build, compared to the 0.91 qmake build. I am assuming this is
because Qt5Gui and Qt5Sql pick them up, but why was this not the case
with 0.91?
- On both of my -current machines, both 0.100 AND the present 0.91p0
display a redraw bug when inserting a library element: the cursor
'guidelines' do not disappear until a redraw is forced by a tiny zoom
in/out with the mouse wheel.
This effect is NOT present on either the old or new 0.100 builds on my
7.9-stable machine.
It looks like a DRM issue to me. Have there been changes to Qt5 graphics
rendering since 7.9-release that could cause this?
Regards
Chris
Further testing shows the redraw bug is not present under cwm or dwm. It
seems to be an XFCE-specific issue.
Chris
