Improve TFT/GUI upload reliability#364
Conversation
- Replace the per-chunk upload loop with a single currentOffset variable that only advances after a valid ACK, fixing silent corruption when the display rejects a chunk (previously the offset advanced before reading the ACK) - Handle 0x08 skip response correctly: non-zero offset jumps ahead in the file; zero offset means "continue sequentially" and now correctly advances currentOffset rather than re-sending the previous chunk (which was corrupting the flash) - Parse all 4 offset bytes for 0x08 with a generous 10s timeout, matching the first-byte timeout — the display sends 0x08 immediately but may take several seconds to compute and transmit the offset - Send connect twice during init; some displays silently discard the first command after DRAKJHSUYDGBNCJHGJKSHBDN, and the 10s read window catches whichever connect gets a response. A 500ms discard read drains any second comok from panels that respond to both. - Always initiate TFT handshake at 9600 baud (Nextion power-on default); switch to tft_upload_baud for data transfer via whmi-wri(s). Retry at 9600 on comok failure. - Fix HttpLib::DownloadChunk: remove available()-capped readBytes call that returned short reads whenever TCP segments were still in flight - Add streaming MD5 over all sent bytes, compared against server checksum after upload completes; comparison skipped if display requested offset jumps - Add consecutive_errors counter with reboot threshold across all ACK failure paths; reset to zero on any successful ACK - Guard tft_update MQTT command: reject with error log if manager_address is not yet known (avoids URL-with-empty-host failure) - Wake display with sleep=0 after comok exchange before sending whmi-wri - Increase NSPanel::init() display power-on wait from 1s to 3s to give the display time to boot before the NSPM flag detection window opens Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
jamesmulcahy
changed the title
|
Hey, this is all awesome work and I really appreciate it! Where did you find all the details in how the panel handles things internally? The firmware has moved from here to it's own repo available here. I can manually move the changes over if you'd like or you can open a PR towards that repo if you care to have your profile considered a contributor to the project. |
|
I asked Claude Code to compare my implementation to the new firmware repo, and work out which changes needed moving. Seems like many of the fixes are already in the newer releases -- which indeed explains why I had so much success updating my panels overnight. It did find one buffer overflow bug, and there are some small tweaks we can port forward -- but the core protocol changes seem to already be incorporated. Attaching Claude's analysis here for visibility -- but I'll submit PRs to the new repo. |
|
Had a read through your attached document and agree. There are a few things that for sure can be handled better. Hopefully this can resolve #340 |
|
I believe this PR can be closed as all the issues raised has been handled in the following PRs:
Do you have any objections? Once again, thanks for your help. |
|
Yep, agreed - I've spun of PRs for anything that didn't look like it was already covered. |
2 participants
Replace the per-chunk upload loop with a single currentOffset variable that only advances after a valid ACK, fixing silent corruption when the display rejects a chunk (previously the offset advanced before reading the ACK)
Handle 0x08 skip response correctly: non-zero offset jumps ahead in the file; zero offset means "continue sequentially" and now correctly advances currentOffset rather than re-sending the previous chunk (which was corrupting the flash)
Parse all 4 offset bytes for 0x08 with a generous 10s timeout, matching the first-byte timeout — the display sends 0x08 immediately but may take several seconds to compute and transmit the offset
Send connect twice during init; some displays silently discard the first command after DRAKJHSUYDGBNCJHGJKSHBDN, and the 10s read window catches whichever connect gets a response. A 500ms discard read drains any second comok from panels that respond to both.
Always initiate TFT handshake at 9600 baud (Nextion power-on default); switch to tft_upload_baud for data transfer via whmi-wri(s). Retry at 9600 on comok failure.
Fix HttpLib::DownloadChunk: remove available()-capped readBytes call that returned short reads whenever TCP segments were still in flight
Add streaming MD5 over all sent bytes, compared against server checksum after upload completes; comparison skipped if display requested offset jumps
Add consecutive_errors counter with reboot threshold across all ACK failure paths; reset to zero on any successful ACK
Guard tft_update MQTT command: reject with error log if manager_address is not yet known (avoids URL-with-empty-host failure)
Wake display with sleep=0 after comok exchange before sending whmi-wri
Increase NSPanel::init() display power-on wait from 1s to 3s to give the display time to boot before the NSPM flag detection window opens