GUACAMOLE-377: Add libguac "guac_display" API for easy and efficient rendering. #525
Commits on Sep 28, 2024
-
GUACAMOLE-377: Establish public guac_display API for efficient render…
…ing via Guacamole protocol.
-
GUACAMOLE-377: Tighten dirty rects down to the individual pixel, rath…
…er than whole 64-pixel lines.
-
GUACAMOLE-377: Correct logic and lock order around deferred frame flu…
…shing (caused deadlock).
-
-
GUACAMOLE-377: Flush an automatic frame for mouse changes only if the…
…re are no other graphical changes pending.
-
GUACAMOLE-377: Updates to cursor must occur within the boundary of th…
…eir relevant frame (before "sync").
-
GUACAMOLE-377: Ensure all layers supporting transparency are properly…
… cleared before receiving data that also supports transparency (including WebP).
-
GUACAMOLE-377: Add ability to "hint" that a drawing operation copied …
…data from another layer.
-
GUACAMOLE-377: Do not use threaded WebP encoder given that guac_displ…
…ay will already make use of a worker thread pool.
-
-
GUACAMOLE-377: Group all guac_display documentation together under a …
…shared Doxygen topic.
-
-
GUACAMOLE-377: Replace RDP settings string (rather than reuse storage…
… of unknown size).
-
GUACAMOLE-377: Use FreeRDP's built-in GDI handling of bitmap cache.
With FreeRDP's GDI being used for ther other aspects of RDP drawing, using our own implementation of bitmap caching causes the FreeRDP's GDI to become out-of-sync with Guacamole's representation, producing graphical artifacts. We can't simply monkey-patch the GDI, as the functions used are internal and not part of the public FreeRDP API. There are likely other possible approaches, like manually updating FreeRDP's GDI surface in addition to Guacamole's surface, but this may not be worth the effort given that bitmap caching is not commonly used by modern RDP servers.
-
-
GUACAMOLE-377: Work around libssh2 usage of strlen() on key data (oth…
…erwise encumbering testing of terminal emulator refactor).
-
GUACAMOLE-377: Increase maximum number of streams.
The new guac_display otherwise tends to run out of outbound, client-wide streams.
-
GUACAMOLE-377: Correct return value confusion in handling of timeout …
…vs. general failure of WaitForMultipleObjects(). The conversion of WAIT_FAILED to a signed int and back may cause the value of result to not actually match the value of the WAIT_FAILED macro due to the difference in size and sign extension during conversion.
-
GUACAMOLE-377: Prefer GUAC_COMP_OVER to GUAC_COMP_SRC for performance…
…-critical operations (~3x faster).
-
GUACAMOLE-377: Remove all old surface, display, cursor, and rect code…
… from src/common/.
-
GUACAMOLE-377: Ensure frame boundaries are sent even for frames conta…
…ining no graphics.
-
GUACAMOLE-377: Allow the pending frame buffer for a guac_display_laye…
…r to be external.
-
GUACAMOLE-377: Allow external guac_display buffers to be replaced wit…
…h NULL if necessary for external cleanup tasks.
-
GUACAMOLE-377: Wrap all VNC message handling within drawing context.
It is otherwise difficult to guarantee that all operations touching the pending frame buffer will occur while holding an open raw context, resulting in unstable behavior.
-
GUACAMOLE-377: Remove unnecessary locking of last frame during resize.
No longer necessary now that the last and pending frame buffers are not interleaved.
-
GUACAMOLE-377: Restore good ol' guac_common_surface and guac_common_c…
…ursor for use by terminal.
-
GUACAMOLE-377: Make vertical combination more likely by limiting comb…
…inations to aligned boundaries.
-
GUACAMOLE-377: Read lock should NOT be acquired at same time as write…
… lock. Acquiring the read lock first and then reentrantly acquiring the write lock may result in other existing readers getting promoted to writers.
-
GUACAMOLE-377: Do NOT combine display operations vertically unless ed…
…ges match exactly. Doing otherwise can result in display operations overlapping. This is because combining two vertically adjacent operations that have different widths causes additional cells to be included that are not covered by either original operation. If other operations are within those additional cells, then the operations flushed for the frame will overlap, consuming unnecessary additional processing and bandwidth.
-
GUACAMOLE-377: Do not allow "img" instructions to break "rect" and "c…
…fill" pairs. Other instructions that occur between "rect" and "cfill" may disrupt the path set by "rect", resulting in "cfill" having no effect, resulting in graphical artifacts. This also has the side effect of reducing thread contention by flushing the simple operations early (it is now less likely that multiple worker threads will be free for further tasks at nearly exactly the same time).
-
GUACAMOLE-377: Reduce number of worker threads to one per processor.
Testing doesn't seem to support an increase in throughput or responsiveness from doubling up on worker threads.
-
GUACAMOLE-377: Add convenient, default render thread implementation f…
…or guac_display.
-
-
-
GUACAMOLE-377: Fix regression causing RDP connections to always recon…
…nect after closing.
-
GUACAMOLE-377: Stream/object sentinel must be assigned BEFORE returni…
…ng index to pool. Doing otherwise results in a race condition where the index of a valid stream is changed to -1 by a different thread, breaking assertions and causing the connection to disconnect.
