mirror of
https://github.com/storytold/photocraft.git
synced 2026-11-21 09:26:43 +00:00
On Windows 11 (winspool.drv 10.0.26100.9444) the custom_print_paper test binary died with STATUS_ACCESS_VIOLATION. Its three tests open the Print dialog at once, and each one lists printers: file.printers calls craft_print::printers(), which opens each printer's PrintTicket provider (PTOpenProvider). winspool creates a process-wide cache without a lock the first time PTOpenProvider is called for a printer whose provider is PrintConfig.dll. When two threads do this together, winspool can leave its pointer to the cache on a freed object, and the next caller enters that object's deleted critical section. The app can make the same calls from two threads: the Print dialog on the UI thread and a print job's worker. print_cmds::with_spooler holds a process-wide lock around every call PhotoCraft makes into craft-print that lists printers or reaches the spooler or a driver: craft_print::printers and submit, and rgb's printers, default_printer, page, print and settings_dialog. A panic inside the lock does not block later callers. While a print job is being handed to the spooler, File › Print waits for it before the dialog opens. Session::printer_source supplies the printers that file.printers, file.print.printers and the Print dialog list, in place of the spooler's. The tests that open Print or list printers set it, so they no longer query the system's printers or load their drivers. print_cmds::system_printer_listings counts the real listings, and the dialog and print job tests assert that it stays 0. The app leaves printer_source unset and lists the spooler's printers as before. New tests: printer_listings_enter_the_spooler_one_thread_at_a_time (eight threads list printers at once; it fails without the lock) and a_panic_inside_the_spooler_does_not_block_later_callers.