Repository navigation
Feat: hello-awtfree draws Compose through the GraalVM window's own surface - #58
Merged
Merged
Conversation
…rface The one-executable gate's sample is now a Compose application with no AWT, built from #42: text, a field and a clickable control, hosted by the core fork's ComposeWindowHost in the GraalVM window and drawn through appKitWindowSurface (Metal) or x11WindowSurface (GL), from core JetBrains#138. With HELLO_SELF_CHECK it still checks Korean with no ICU data file and the inked drawing, and now also presents 30 frames of the Compose content, none at a size other than the window's, which check-one-executable.sh requires. A run that does not end within 120 s is stopped with every thread's stack printed (sample on macOS, gdb on Linux). The GraalVM macOS cell hung after its text check once on #57 (run 39058147045, attempt 1, between drawing the PNG and the window report); the next one names where. The window modules are built with the scene, against the core fork's Compose, which the window module jobs now take first. The core fork is pinned at 8130daf5, which has the surfaces.
Run 39067773624: with the scene, checkNativeImageNoAwt fails on Linux (libawt.so, libawt_xawt.so, libfontmanager.so, libjavajpeg.so and liblcms.so written beside the image) and on macOS (Java_sun_font_, libawt, libjawt, libosxui and libfontmanager named inside it), while none of the forbidden root types was reached. The types whose loading pulls in AWT's natives without a window (the font manager, Color, Font, Cursor, ColorModel, Clipboard) are now forbidden too, so the analysis names the path; and, for this PR only, the call tree is written and uploaded.
…arm64 The forbidden AWT types were passed as one comma-joined -H:ReportAnalysisForbiddenType value (since #52), which the option does not split, so nothing was forbidden: run 39070735944's macOS call tree has UIManager reached from com.apple.laf class initialisers with no report. Each type now has its own option, so the analysis stops and names the path. Linux arm64's link failed on eglGetProcAddress, eglQueryString and eglGetCurrentDisplay once the scene made Skia's GL backend reachable: JetBrains' Skia for that platform finds GL through EGL.
… images With the forbidden types working, run 39073283082 stopped on java.awt.Toolkit reached from androidx.compose.ui.input.key.Key.toString (KeyEvent.getKeyText). The core fork guards that, the system clipboard, the Transferable clipboard fallback and skiko's Swing dispatcher behind ExtendedAwt.available, read from -Dcompose.awt=false, but only a class initialised while the image is built makes the guard a constant. The plugin passed the property and left the class to run time, so every guarded branch stayed reachable. ExtendedAwt is now initialised at build time alongside AwtSwitch whenever it is on the classpath, as compose-rust's renderer does.
Run 39074824167 builds hello-awtfree with no AWT on all three cells; the only thing left beside the executable was the call tree's reports folder this PR turned on to find the path. The sample keeps NATIVE_IMAGE_CALL_TREE for the next time a path has to be found.
Run 39079050531 caught the macOS hang with the new thread dump. The process main thread is in _pthread_join, and another thread waits in dispatch_sync for the main queue (the AppKit window's dxc_on_main), which nothing serves, so it never returns. Every frame in the image was unnamed; for this PR the image keeps its local symbols, so the next sample names them. On Linux x64 and arm64 the scene came up and presented a frame with 0 at the wrong size. It presented only one because nothing in the content changes; the self-check now asks for a frame each turn, so 30 are drawn and presented.
…en main is off the first thread The symbolised sample of the macOS hang (run 39082357989) shows it: the process's first thread is in JavaMainWrapper.doRunInNewThread, joining the thread main runs on, while that thread waits in dxc_native_frame_end for the main queue, which only the first thread serves. The sample gave -H:-RunMainInNewThread after the plugin had locked experimental options again; it now unlocks them first. The self-check also asks pthread_main_np before opening the window and exits with a message instead of hanging, so whether main is on the first thread is a line in the report.
… runs main on the first thread Run 39084384117 settled the macOS hang: with -H:-RunMainInNewThread given and unlocked, main still reports pthread_main_np() == 0 under Liberica NIK 25.0.4.1, and the symbolised sample before it has the first thread in JavaMainWrapper.doRunInNewThread joining main while main waits in dispatch_sync for the main queue. NIK runs main on a new thread on macOS whatever the option says. The AWT-free samples use nothing of NIK's, so the macOS cell takes upstream GraalVM 25.0.1, which honours the option. This is also the intermittent hang of hello-awtfree-link: it stops at the first AppKit call that crosses threads, and that call is not on every path.
…stop keeping symbols in the image The macOS hang is traced and fixed; the image goes back to deleting its local symbols.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Supersedes #42 and answers thisisthepy/python-multiplatform#191 §5 item 2 (a demo that draws Compose on the GraalVM path).
hello-awtfreeis now a Compose application with no AWT. It uses Feat: native-image hello-awtfree, a window with no ComposeWindow and no java.awt #42's content: Korean text, a field and a clickable control. The core fork'sComposeWindowHosthosts it in the GraalVM window, drawn throughappKitWindowSurface(Metal) orx11WindowSurface(GL) from core Scrollbars do not appear at all on MacOS 11.0.1 JetBrains/compose-multiplatform#138.HELLO_SELF_CHECK, the sample keeps the Korean check with no ICU data file and the inked drawing. It now also presents 30 frames of Compose content, none at a size other than the window's.check-one-executable.shrequires that line.sampleon macOS,gdbon Linux). The macOS cell hung once on Chore: Publish the plugin as org.thisisthepy.compose and resolve the core fork's libraries #57 (run 39058147045, attempt 1), after the PNG and before the window report. The next hang will show where it waits.scene, against the core fork's Compose.CORE_EXTENDED_REFmoves to 8130daf5, which has the surfaces.