FreeType’s FT_Glyph_To_Bitmap mutates the FT_Glyph by allocating a new
glyph object and overwriting the input pointer. This creates a risk of
memory leaks if the old glyph isn’t released or if ownership is unclear.
To avoid this, we now store the glyph in a UniqueFTGlyph and explicitly
pass it by reference to FT_Glyph_To_Bitmap. After the mutation,
ownership continues to reside in the smart pointer, and cleanup is
guaranteed even in error branches.
For temporary glyphs (e.g., strokes), we still manually call
FT_Done_Glyph, since they are not managed by UniqueFTGlyph. These
changes ensure all glyph memory is correctly released and reduce the
need for repetitive error-path cleanup code.
This fixes the ability for users with uppercase letters in their
username to change their passwords, which wasn't possible before on
non-Windows platforms. The underlying issue for that is
https://github.com/processone/ejabberd/issues/4377 and in addition some
inconsistent normalization of usernames in password change requests
by gloox. This commit works around that by always using the local JID
part as username for password requests, which got the nodeprep string
profile already applied.
It also fixes a problem that Windows users which were able to change
their passwords, weren't able to login afterwards anymore, unless they
typed their username in all lowercase in the login form. This was caused
by using the all lowercase username as input for the password hash
function, instead of using the username in the user supplied case.
Fixes#7796
If a commit just removes files for cleanup purpose for example the list
of files to check would be empty. Therefore allow no files as arguments
for use in CI.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
A last minute change in #7787 caused the bold and italic identifiers of
fonts to be ignored once it got merged in 734386ce9f and all fonts to
be rendered as regular fonts. This commit fixes that.
The perl script extract.pl just checks for a line matching
'-- include data json--', so making it valid js is can be done just
fine.
Also fix eslint warning is graph.js.
Ref: #7812
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Replace premade bitmap fonts with FreeType2 and dynamic texture atlas.
Switched from static, premade bitmap fonts to runtime-generated glyphs using FreeType2.
Glyphs are rendered on demand into a dynamic texture atlas, improving scalability,
font quality, and support for internationalization.
Currently using RGBA format for compatibility across all render
paths while initial support for R8_UNORM is being added.
This prepares for future optimizations in VRAM and performance
Includes groundwork for gamma-corrected blending and future swizzling support.
The remaining ones that can be fixed automatically.
eslint --no-config-lookup --fix --rule '"prefer-const": 1'
Ref: #7812
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
This fixes all issues eslint can safely do automatically. Drastically
reducing the number of open style issues reported.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
This allows selecting multiple reports to load at once in the file
chooser dialog. Reports will be loaded in parallel and the last one to
finish loading will be the active one.
This removes a duplicate call of on_report_loaded() when loading a
profiling report. This second call caused loading reports to fail,
whenever it took longer than 5 seconds. While that was probably a
measure to prevent reports loading too long and taking up too much
resources, it didn't achieve that goal, as the actual loading of the
report din't get aborted.
This replaces the previous arclint linter for checking the copyright
year in license headers with a gitea workflow job.
As the date of last edit might differ from commit date due to reverts
the copyright linter is run against a base commit.
The python script doing the heavy lifting is written by @Dunedan.
Signed-off-by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Co-Authored-by: Dunedan <dunedan@phoenitydawn.de>
Adds possibility to control the frame time by setting the frame
frequency by a command line argument (`-fixed-frame-frequency`). It's
mostly useful for a trailer recording and graphics debugging.
It allows to reduce the number of allocations per-frame for
ShaderDefines. For example, on Acropolis Bay it goes down from
2.7k to 0.9k for the whole map view.
We don't really need std::vector for it. Because materials and
particularly shaders shouldn't have so many parameters.
Shrinking GCs dump the JITted code, which leads to redundant recompilations, lowers performance, and makes profiling JS more difficult.
They may still happen if the runtime is at risk of OOM.
According to our stats `GL_ARB_fragment_program_shadow` is supported
by all our GL users. Also we're going to remove OpenGL ARB in A29. So
I remove ARBShadersShadow completely.
A multi player loading screen will have possibility of showing single
and multi player tips while multi player tips are in favor.
Tips are now separated by categories. During loading those category
have an occurrence probability, where we are choosing a random category
based on chance and not by complete randomness. From there we just
picking a random tip from that category.
SpiderMonkey 98 introduced a size heuristic for the nursery GC region
(https://phabricator.services.mozilla.com/D136637). As this heuristic
uses a wall-clock time duration, it results in a severe performance
regression on slower systems for our use case.
This commit adds a workaround to turn off that heuristic, by telling
SpiderMonkey that a "page load" (something which doesn't have a meaning
in the context of pyrogenesis) is in progress, as that heuristic is
disabled for page loads.
Co-Authored by: Ralph Sennhauser <ralph.sennhauser@gmail.com>
Fixes#7714
Since 0eed117e6d exceptions from the map generation script are rethrown
in the main thread but not all of them are catched.
This defect is already fixed in the engine with 82513c9104Fixes: #7620
Profiler2 currently explicitly considers 'frame' for frames, which is a bit annoying in replays as state hashes are computed outside of these. It's better to just rely on RecordFrameStart() calls.
trompetin17