Repository navigation
Tags
Tags
Release workflow: make test's oracles installed (poppler-utils, libop… …enjp2-tools, qpdf) 2.36's first Release run stopped in make test: tests/jbig2.py and tests/onebit.py check the decoders against poppler's pdfimages and pdftoppm, tests/jpx.py against OpenJPEG's opj_decompress (#48), and the runner had neither. qpdf for tests/crypt.py too. Oracles only: nothing of them is linked or shipped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.35 EPDF draws View > Pages reliably on a 4 MB Falcon. - On a Falcon with 4 MB and NVDI, View > Pages could stay blank: the page renderer (render.ovl, 643 KB) found no piece of memory big enough once the document had been read. EPDF now sets that memory aside when it opens a document, and gives it back at once if the text or a picture needs it (3ac2874). - render.ovl is 9 KB smaller (no GIF decoder: a PDF holds no GIF). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.34 EPDF and EConvert answer the user sooner while they work (phase 3a). - EPDF, View > Pages: pages passed over aren't drawn. While a key is held, or pages are paged through quickly, nothing is drawn; once the view has been still for a moment, only the page it stopped on is drawn. A single move (a click in the page strip, Find, one key) draws at once (#96, e8f115b3). - EConvert stays responsive while it converts: its longest stretch without answering on the showcase is down from 160 ms to 50 ms, and the converted files are byte for byte as 2.33's (d6f58de). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.33 The page strip and the pages' previews (Erez, 15 Nov 2026: "start the page strip after 2.32 and complete the functionality"; then "accept and make it a setting flag, off by default. View page strip off by default"). - View > Page strip (off by default): the pages in a column at the window's left, each its preview where the PDF has one, on a Falcon or TT in View > Pages the page drawn small, else a box with its first line; a click goes to the page (de28eec). - EConvert's Options > Settings... > General > Page previews (off by default): each converted page carries a small picture of itself for the strip, about 2 s a page more converting on a Falcon (eb9d8b8). - Through the Gateway the previews are asked only with Page previews ticked, or EPDF's Open on the Gateway with the strip shown: AERA-Gateway 1.48 (gateway/ at 05af1618). Checked here, on Erez's TOS 1.04 (ST) and 4.04 with NVDI (Falcon-4:256): the round at eb9d8b8 (every rig; econv_speed_show's known row 160 of 50), speed.py against 2.32 side by side within a tick or 2.32's spread, CONVTEST without previews 249,420 ms (2.32: 249,230, the file the same), with them 266,840; gateway.py against AERA-Gateway 1.48 with and without GW_PREVIEWS (0dda5ff); tools/rpad.py 0 on every hot function, EPDF.PRG's code at 2.32's places mod 256. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.32 A fault fix: on a FireBee (FireTOS: _CF_, _CPU 60) EPDF loaded render.ovl, 68030 code a ColdFire doesn't run (on common's stand-in, a 68060 Falcon with FireTOS's cookies, exception 61 on the first page). Now a FireBee keeps to the text view and a converted copy's stored pages, as an ST, and EConvert converts there through the Gateway only (65c3b65, common b4d9cf7). Checked here, on Erez's TOS 1.04 (ST) and 4.04 with NVDI (Falcon-4:256, his configuration): - rovl.py and gui.py on FireBee:16 (a stand-in, not a FireBee): render.ovl not loaded, said why; gui ok; - the round: rovl, convtest, speed, gui, files, pi, gateway, translate, econv, econv_fsel, econv_speed, zoom, drawn, prepared pass; econv_speed_show's "Convert: its turns" 155 of 50 (2.31: 165); - CONVTEST: 249,230 ms, as 2.31; - speed.py against 2.31 side by side, TOS 1.04 / 4.04: ST:mono and ST:med all within on both; Falcon-4:256 the same five rows over as 2.31 (a line while reading 75/75, a line 70/70, a line up 75/80, Select all 75/70, About 205/210); every row within +-10 but ST:mono's Select all 55 -> 40 and Falcon-4:16's next page 90 -> 105 (of 300); - tools/rpad.py against 2.30: 0 on every hot function of render.ovl, econvert.elf and convtest.elf. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.31 A fix forward: on a Vampire V4SA with its new Atari core EPDF hung at its start (EWeb and EPlay too; EMail and EHypView were ended). common's NatFeats probe put its handler at 0x10 and ran an illegal instruction; the core keeps VBR away from 0. Now EPDF and EConvert look for NatFeats only when their INF says NATFEATS=1, which only the test rigs write (Erez, 15 Nov 2026), and nf.S finds VBR itself (common e6c65c3); with c9a6896, conn.c's log lines written at once. Checked here, on Erez's TOS 1.04 (ST) and 4.04 with NVDI (Falcon-4:256, his configuration): - with no NATFEATS line EPDF opens its window and reads its document with nothing on NatFeats (no probe); with it, it logs as before; - the round: rovl, convtest, gui, files, pi, gateway, translate, econv, econv_fsel, econv_speed, zoom, drawn, prepared pass; econv_speed_show's "Convert: its turns" 165 of 50 as in 2.30 (165); - rovl.py, Falcon-4:256: every page's longest stretch 45-50 ms in all three modes, page times within 15 ms of 2.30's; - CONVTEST: the file byte for byte 2.30's, 249,495 -> 249,230 ms; - speed.py against 2.30 side by side, TOS 1.04 / 4.04: ST:mono and ST:med all within their times on both; Falcon-4:256 the same five rows over as 2.30; every row within +-10 but ST:med's "the start" (145 -> 160, run again: 2.30 160, 2.31 155); - tools/rpad.py against 2.30: 0 on every hot function of render.ovl, econvert.elf and convtest.elf; EPDF.PRG's drawing, text, libc and libgcc at 2.30's places mod 256 (EP_PAD_A..C, EC_PAD 92, CT_PAD 172). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.30 Phase 3a: a page drawn on a Falcon or TT gives the machine back every 50 ms or less, the progressive photo's page too (Erez, 15 Nov: "go with the tick every 8"). - common 89b5e5e: libjpeg's AERA_TICK every 8 MCUs in jdcoefct.c (was every 16): the showcase's page 3, 70 -> 45-50 ms, +80-90 ms a page. - placement kept: LJ_PAD 42 -> 52; tools/rpad.py 0 on every hot function of the three 68030 programs against 2.29. rovl.py, Falcon-4:256 (TOS 4.04, NVDI), longest stretch per page, 2.29 -> 2.30: DSP 50 45 70 50 50 50 45 45 45 -> 50 45 45 45 45 45 50 50 45; 68030 50 45 70 50 45 50 45 50 45 -> 50 45 50 45 45 45 45 45 45; bands 50 50 70 50 50 45 50 50 45 -> 50 50 50 50 50 50 45 50 50. Page 3 +80-90 ms, the others within 15. CONVTEST: the file byte for byte 2.29's, 249,445 -> 249,495 ms (page 3 +55). speed.py against 2.29 side by side: every row within +-10 on ST:mono, ST:med, Falcon-4:16 and Falcon-4:256 (Falcon-4:256's line while reading 80, 70-80 in every round since 2.25), the same rows over as 2.29. The round: gui, files, pi, gateway, translate, econv, econv_fsel, econv_speed, zoom, drawn, prepared pass; speed's ST:mono "a line scrolled" 55 of 50 (2.29 the same) and econv_speed_show's "Convert: its turns" 165 of 50 (160-170 since 2.25) as before. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.29 Phase 3a: a page drawn on a Falcon or TT gives the machine back every 50 ms or less, but for a progressive photo's 70 (the showcase's page 3). - rc_fill turns by its work (pixels across, crossings, edges made and sorted), one count for every fill: page 5's 60 -> 45 ms, its turns 545 -> 200; in bands 80 -> 50. - a stream's stored bytes read and copied 32 KB at a time with the reader's stop hook between; - a turn after a JPEG's headers (common 9d36370): page 2's 60 -> 45. - placement kept: APAD_F 116, APAD_B 118, STREAM_PAD 6, PIC_PAD 134; tools/rpad.py 0 on every hot function against 2.28. rovl.py, Falcon-4:256 (TOS 4.04, NVDI), longest stretch per page, 2.28 -> 2.29: DSP 45 60 70 50 60 45 45 50 50 -> 50 45 70 50 50 50 45 45 50; 68030 45 55 70 45 60 50 50 45 45 -> 50 45 70 50 45 50 45 45 45; bands 50 55 70 50 80 50 50 45 45 -> 50 50 70 50 50 45 50 50 50. Page times within 15 ms (bands p2 +40). CONVTEST: the file byte for byte 2.28's, 249,205 -> 249,445 ms, page 1 +95 (Erez, 15 Nov: accepted). speed.py against 2.28 side by side: every row within +-10 on ST:mono, ST:med, Falcon-4:16 and Falcon-4:256; the round's speed rig with no row over on the STs and Falcon-4:16. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
Release 2.28
ST medium's text an unrolled cell at a time (common's cells_st_med),
the window's scroll through common's planar scr_rows.
The speed rig, 2.27 -> 2.28 side by side (longest stretch, ms; EmuTOS
1.4 on the STs, TOS 4.04 on the Falcons):
ST:med a page scrolled 160 -> 115, while reading 170 -> 120,
next page 255 -> 200, the last page 195 -> 150, the start
200 -> 150, View > The page's lines 270 -> 225, About 175 ->
165: all within their times (2.27: 2 over)
ST:mono three runs each, every row within 2.27's own spread
Falcon-4:16 every row within a tick (next page 100 -> 105, a page
scrolled 70 -> 75, the start 95 -> 95)
Falcon-4:256 every row within a tick, the 5 rows over as in 2.27
CONVTEST (Falcon-4:256, the showcase): the file byte for byte 2.27's,
249.14 -> 249.21 s. The round: rovl, gui, files, pi, gateway,
translate, econv_speed, econv_fsel, econv, zoom, drawn, prepared pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SZeNoDLt48EM4msuao4J1n
PreviousNext