*Added global connectme flag that allows for irregular/unusual objects to be properly positioned when being autopatched. This also deals with objects that do not have their top-left corner aligned with the (0,0) coords.

Improve bendin compatibility with vanilla.

See merge request jwilkes/purr-data!489

Correct some errors and add a note about vanilla compatibility options.

This adds an option to make bendin use the vanilla range of 0..16383 for
compatibility. As was observed a long time ago, this is at odds with the
bendout range of -8192..8191, and was thus fixed in pd-l2ork at some
point. But Miller Puckette has decided that this bug won't *ever* be fixed in
vanilla, and this is even documented in vanilla's midi-help.pd patch.

This discrepancy breaks a lot of patches involving MIDI, so a compatibility
option is needed. It's been possible to use -legacy to get vanilla's bendin,
but it's better not to conflate legacy iemgui positioning issues with the
bendin implementation. The former won't usually break interoperability, but
the latter does.

So, as suggested by Ico Bukvic, we add a flag as an optional second argument
to bendin instead. If present, this option explicitly denotes the target
range: zero means the default (pd-l2ork, signed) output range, nonzero the
vanilla-compatible unsigned range. In addition, there's a new -legacy-bendin
command line option which makes unsigned output the default for bendin objects
which don't specify the range explicitly. You'll want to put this into your
startup flags in the preferences if you need out-of-the-box vanilla
compatibility.

Disable nwworkingdir workaround in gui_canvas_saveas for nw.js 0.46+, fixes the first bug reported in #669.

Additional optimizations to avoid unnecessary runtime string comparisons for the purpose of OS detection.

Fixed inconsistent window sizes across Linux and OSX (this was a bug), and Linux/OSX and Windows (this is an nw.js inconsistency linked to innerHeight and innerWidth misreporting their size for some reason on Win10).

Fixed the window resizing bug when saving a patch with optimal zoom, closing it, and then reopening it (see http://disis.music.vt.edu/pipermail/l2ork-dev/2020-August/002901.html). Please note that this will need to be reverted once we migrate to nw.js 0.46+ (as per provided comment inside the canvas_check_geometry. Also improved the patch scrollbars to be flush against the window edges.

Instead, we just test for object creation.

This currently misses cases where abstractions can't create their dependencies
from other abstraction libs, plus some other cases. But as long as we aren't
testing for that, we don't need the noise of the error messages.

In my test, this reduced log size from 4.2MB to 3.2MB.

Example of the new output (taken from Gem compilation):

      CXX      libplugins_la-record.lo
    record.cpp:58:3: warning: #warning better handling of duplicate codecs [-Wcpp]
     # warning better handling of duplicate codecs
       ^~~~~~~
    record.cpp:87:3: warning: #warning isThreadable [-Wcpp]
     # warning isThreadable
       ^~~~~~~
    record.cpp:257:3: warning: #warning enumerateProperties [-Wcpp]
     # warning enumerateProperties
       ^~~~~~~
      CXX      libplugins_la-video.lo
      CXX      libplugins_la-videoBase.lo
      CXXLD    libplugins.la

Not all components honour the V=0 flag -- perhaps they don't use
Automake, or use a very old version. However, many do.

See also: https://www.gnu.org/software/automake/manual/html_node/Automake-Silent-Rules.html

See also: jwilkes/purr-data#666

Jean-Michaël Celerier authored 4 years ago and Albert Gräf committed 4 years ago

Else this does not seem to build with MSVC 2019

The GitLab CI commands are now separated from the job definitions, using
YAML anchors.

The build and test commands should now show as separate, collapsible
sections in the job output (as documented at
https://docs.gitlab.com/ee/ci/pipelines/#custom-collapsible-sections).

The Debian image preparation is moved to before_script to separate it
from the build and test commands. Note that this adds an implicit 5 minute
time limit on this part of the script (see
https://gitlab.com/gitlab-org/gitlab-runner/-/issues/2716).

The commands are also now written once and copied using YAML anchors
(as documented at
https://docs.gitlab.com/ee/ci/yaml/README.html#yaml-anchors-for-before_script-and-after_script).

Addressed the issue where optimal zoom did not update the scrollable area and reenabled the zoom of the scrollbars when zooming (this scrollbar zoom size may require further reflection to decide the best path forward).