@geastack/cli 0.1.115 → 0.1.116
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/docs/SETUP.md +20 -12
- package/package.json +2 -2
- package/src/esp32/build.mjs +19 -19
package/docs/SETUP.md
CHANGED
|
@@ -336,7 +336,8 @@ at call time, so never record a port path.
|
|
|
336
336
|
### Composed targets
|
|
337
337
|
|
|
338
338
|
A composed target is a JSON definition that extends a base (`esp32-s3` or
|
|
339
|
-
`esp32-p4`, with `mcu` set to `esp32s3` or `esp32p4` to match) and picks its
|
|
339
|
+
`esp32-p4`, with `mcu` set to `esp32s3` or `esp32p4` to match) and picks its
|
|
340
|
+
chips from the `@geastack/chips` catalog (`gea chips list`). Every
|
|
340
341
|
role is optional, so the definition describes the hardware that is there:
|
|
341
342
|
`display`, `touch`, `power`, `imu`, `audio`, `rtc` and `expander`. Some boards are
|
|
342
343
|
shipped composed in `@geastack/targets` and need no definition of your own,
|
|
@@ -413,23 +414,30 @@ to one role. The CrowPanel Advance 7.0 definition uses all of these.
|
|
|
413
414
|
|
|
414
415
|
#### The `esp32-p4` base
|
|
415
416
|
|
|
416
|
-
A P4 definition takes the same roles, with GPIOs up to 54.
|
|
417
|
-
|
|
418
|
-
|
|
419
|
-
|
|
420
|
-
|
|
421
|
-
|
|
422
|
-
`flashMode`.
|
|
417
|
+
A P4 definition takes the same roles, with GPIOs up to 54. The base is the
|
|
418
|
+
Waveshare ESP32-P4 7-inch project (`esp32-p4-waveshare-touch-lcd-7`), as
|
|
419
|
+
`esp32-s3-touch-amoled-2.06` is for `esp32-s3`; everything the definition
|
|
420
|
+
names replaces that board's own chips, console, camera sensor and wireless
|
|
421
|
+
wiring. Its sdkconfig and partition table are laid out for the Waveshare's
|
|
422
|
+
32MB flash at DIO 40MHz, so a board with other flash sets `flashSize`,
|
|
423
|
+
`flashMode`, `flashFreq` and its own `partitions`.
|
|
424
|
+
|
|
425
|
+
Its display is a parallel RGB panel scanned straight out of the P4's
|
|
426
|
+
in-package PSRAM by the LCD driver, through only a small FIFO: a PSRAM stall
|
|
427
|
+
of a few microseconds underruns it and the picture creeps right. Run flash the
|
|
428
|
+
way the vendor's own firmware does; the CrowPanel Advanced 5.0 underran every
|
|
429
|
+
few seconds at DIO 40MHz and never at its vendor's QIO 80MHz. Use
|
|
423
430
|
`scanFramebuffers: 3` so the renderer draws straight into a buffer not being
|
|
424
431
|
scanned (one buffer adds a PSRAM-to-PSRAM copy of every changed region; with
|
|
425
432
|
two, each swap can wait up to a frame for the panel to let go of the old
|
|
426
433
|
front). Scroll copies and full-width patches between the buffers go through
|
|
427
434
|
the AXI-GDMA, about three times faster than the CPU. Keep the L2 cache's
|
|
428
435
|
64-byte lines: 128-byte lines speed CPU copies up but stall PSRAM long enough
|
|
429
|
-
for the panel to underrun at 60Hz. Leave `restartInVsync` off; the driver
|
|
430
|
-
if it ever falls behind. `psram` is
|
|
431
|
-
it up). `pclkIdleHigh: true` idles the pixel
|
|
432
|
-
some panels need. Two sections exist only on
|
|
436
|
+
for the panel to underrun at 60Hz. Leave `restartInVsync` off; the driver
|
|
437
|
+
restarts the DMA itself at the next VSYNC if it ever falls behind. `psram` is
|
|
438
|
+
not configurable (the base sets it up). `pclkIdleHigh: true` idles the pixel
|
|
439
|
+
clock high between lines, which some panels need. Two sections exist only on
|
|
440
|
+
this base:
|
|
433
441
|
|
|
434
442
|
- `wireless`: the WiFi/BLE coprocessor (an ESP32-C6, say) on SDIO through
|
|
435
443
|
esp_hosted: `slot`, `busWidth` (1 or 4), `clockKhz`, `resetActiveHigh`, and
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@geastack/cli",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.116",
|
|
4
4
|
"description": "Command-line front door for GeaStack apps, targets, and local toolchains.",
|
|
5
5
|
"license": "Apache-2.0",
|
|
6
6
|
"repository": {
|
|
@@ -38,7 +38,7 @@
|
|
|
38
38
|
"dependencies": {
|
|
39
39
|
"@clack/prompts": "^1.8.0",
|
|
40
40
|
"@geastack/simulator": "^0.1.15",
|
|
41
|
-
"@geastack/targets": "^0.1.
|
|
41
|
+
"@geastack/targets": "^0.1.113",
|
|
42
42
|
"picocolors": "^1.1.1",
|
|
43
43
|
"serialport": "^13.0.0"
|
|
44
44
|
},
|
package/src/esp32/build.mjs
CHANGED
|
@@ -257,19 +257,6 @@ export function applySdkconfigPolicy(sdkconfig, { selection, app, capabilities,
|
|
|
257
257
|
}
|
|
258
258
|
}
|
|
259
259
|
|
|
260
|
-
// A bare RGB panel is scanned out of PSRAM through bounce buffers refilled
|
|
261
|
-
// by the LCD interrupt. Its code must sit in IRAM: otherwise it is fetched
|
|
262
|
-
// over the same PSRAM bus rendering saturates (always so with XIP from
|
|
263
|
-
// PSRAM), the refill misses its slot, the FIFO underruns and every later
|
|
264
|
-
// line lands shifted -- the picture shreds on each repaint and stays slid.
|
|
265
|
-
// Restarting the transfer at each VSYNC puts a slipped frame back. In that
|
|
266
|
-
// mode the driver never resets the EOF count it picks each refill's buffer
|
|
267
|
-
// from, so one lost EOF would leave every refill a buffer out of step with
|
|
268
|
-
// the DMA (thin lines at each band's head); rgb_panel.cpp's refill reads the
|
|
269
|
-
// buffer and band from the DMA itself instead. A board whose display says
|
|
270
|
-
// `restartInVsync: false` skips the restart: it runs in the VSYNC interrupt,
|
|
271
|
-
// which waits out a refill in progress, and once that pushes it past the
|
|
272
|
-
// vertical blanking the restarted frame lands shifted sideways.
|
|
273
260
|
// A P4's hosted WiFi/BLE coprocessor: esp_hosted takes its SDIO bus and the
|
|
274
261
|
// coprocessor's reset line from these, so they are the board's wiring.
|
|
275
262
|
if (boardWireless) {
|
|
@@ -332,6 +319,19 @@ export function applySdkconfigPolicy(sdkconfig, { selection, app, capabilities,
|
|
|
332
319
|
}
|
|
333
320
|
}
|
|
334
321
|
|
|
322
|
+
// A bare RGB panel is scanned out of PSRAM through bounce buffers refilled
|
|
323
|
+
// by the LCD interrupt. Its code must sit in IRAM: otherwise it is fetched
|
|
324
|
+
// over the same PSRAM bus rendering saturates (always so with XIP from
|
|
325
|
+
// PSRAM), the refill misses its slot, the FIFO underruns and every later
|
|
326
|
+
// line lands shifted -- the picture shreds on each repaint and stays slid.
|
|
327
|
+
// Restarting the transfer at each VSYNC puts a slipped frame back. In that
|
|
328
|
+
// mode the driver never resets the EOF count it picks each refill's buffer
|
|
329
|
+
// from, so one lost EOF would leave every refill a buffer out of step with
|
|
330
|
+
// the DMA (thin lines at each band's head); rgb_panel.cpp's refill reads the
|
|
331
|
+
// buffer and band from the DMA itself instead. A board whose display says
|
|
332
|
+
// `restartInVsync: false` skips the restart: it runs in the VSYNC interrupt,
|
|
333
|
+
// which waits out a refill in progress, and once that pushes it past the
|
|
334
|
+
// vertical blanking the restarted frame lands shifted sideways.
|
|
335
335
|
if (boardRgbPanel) {
|
|
336
336
|
board('CONFIG_LCD_RGB_ISR_IRAM_SAFE', 'y')
|
|
337
337
|
if (boardRgbPanel.restartInVsync) board('CONFIG_LCD_RGB_RESTART_IN_VSYNC', 'y')
|
|
@@ -811,14 +811,11 @@ function runInTarget(command, args, { cwd, env, dryRun, verbose, label, logFile,
|
|
|
811
811
|
return runQuiet(command, args, { cwd, env, dryRun, verbose, label, logFile, stdout, stderr, failureCode })
|
|
812
812
|
}
|
|
813
813
|
|
|
814
|
-
// App capability flags are CMake cache entries. Reconfigure only when those
|
|
815
|
-
// inputs change; ordinary source/CMake dependency changes remain the build
|
|
816
|
-
// system's responsibility. This retains Ninja's sub-second no-op.
|
|
817
814
|
// A build directory belongs to the CMake project it was configured from, and
|
|
818
815
|
// CMake refuses to reuse it for another ("does not match the source used to
|
|
819
|
-
// generate cache").
|
|
820
|
-
//
|
|
821
|
-
//
|
|
816
|
+
// generate cache"). When a target's project changes (a board moved onto another
|
|
817
|
+
// base project, a renamed checkout), its cache goes and configure starts the
|
|
818
|
+
// directory afresh.
|
|
822
819
|
export function dropCacheFromOtherSource(buildDir, sourceDir) {
|
|
823
820
|
const cacheFile = path.join(buildDir, 'CMakeCache.txt')
|
|
824
821
|
if (!existsSync(cacheFile)) return false
|
|
@@ -830,6 +827,9 @@ export function dropCacheFromOtherSource(buildDir, sourceDir) {
|
|
|
830
827
|
return true
|
|
831
828
|
}
|
|
832
829
|
|
|
830
|
+
// App capability flags are CMake cache entries. Reconfigure only when those
|
|
831
|
+
// inputs change; ordinary source/CMake dependency changes remain the build
|
|
832
|
+
// system's responsibility. This retains Ninja's sub-second no-op.
|
|
833
833
|
export async function ensureConfigured({ idf, prepared, env, dryRun = false, verbose = false, stdout, stderr = console.error }) {
|
|
834
834
|
if (!dryRun) dropCacheFromOtherSource(prepared.binaryDir || prepared.buildDir, prepared.targetDir)
|
|
835
835
|
const signatureFile = path.join(prepared.buildDir, '.gea-configure-args')
|