Drop the plain screens the chrome replaced, and test what ships

[oroderico]

seedtool_render_screen3, seedtool_render_screen4 and seedtool_render_list are the text and list screens from before the nav chrome. Every screen that used them was migrated across #11, #12 and #14, and nothing in the firmware has called them since.

Nothing said so. -Werror=unused-function only sees static functions, and these are public — declared in seedtool_render.h and seedtool_display.h, implemented in seedtool_render.c, seedtool_display.c and origo_sdl.c. To the compiler, any translation unit might still call them.

This saves no space

The linker already dropped them. None of the three symbols appears in the built ELF, and the image is 303920 bytes before and after, to the byte.

What it buys is 116 fewer lines for a reader who has to understand the whole tree to trust it — and the test fix below, which is the real reason to do it.

The self-test was the only caller left

Two of its four calls were the point rather than the obstacle. They drew the old list and said they exercised "both arrows and the selection bar". Those arrows are that renderer's scroll indicator; the shipping list uses a thumb. They were proving a widget the firmware had stopped sending — the same failure "Take the tick to the lists and the paged screens too" already found once, from the other side: a check that only knows how to see presence cannot see excess, and one that draws the wrong widget cannot see the right one.

The one that needed care

The other two calls are wire_order_is_big_endian, which pins the panel's byte order — the one property the host can hold the hardware to, and the reason the README says a colour right in the simulator can still be wrong on the board. It paints a list to obtain the selection orange, "the colour that actually came out blue".

Switching its renderer risked leaving the test green over an image without the colour it exists to check. Measured before switching: both renderers put 226 orange pixels on row 21, the row it inspects, with black and white present in both. The test also ends in return orange_found, so a wrong image fails it rather than passing hollowly.

Comments

Seven comments named the removed functions to explain layout constants that are still in use. They are rewritten, not deleted: the four rows still run 28 to 88 in steps of 20, and that still needs a reason — just not one pointing at a function nobody can read any more.

Verification

Host self-test, 52 Python tests, QR smoke test, the self-test under ASan and UBSan, and a firmware build the ELF audit passes at 303920 bytes.

Flashed to a TTGO T-Display, and the Origo menu confirmed still orange on the panel. That is the half of the byte-order question the host cannot answer, and the menu is the screen whose renderer changed.