Handle CSI intermediate bytes instead of printing the final byte

[kay-ws]

Problem

Control sequences containing an intermediate byte have their final byte printed to the screen as a literal character.

printf 'A\e[2 qB\n'

Expected AB, actual AqB. The stray character lands at the cursor position, so an application emitting such a sequence on every cursor movement accumulates it across the line.

Cause

ECMA-48 5.4 defines a control sequence as CSI P...P I...I F, where the Intermediate Bytes I (02/00 to 02/15) "together with the Final Byte F identify the control function".

The state machine has no state for intermediate bytes. On encountering one in EScsi, the lookup in escape_map[] misses and falls through to the default entry of escape_sequences[], which resets the state to ESnormal. The final byte then arrives outside of any sequence and is processed as ordinary text.

This affects every sequence carrying an intermediate byte. ECMA-48 table 4 lists the functions that use a single intermediate byte 02/00 (SL, SR, GSM, ...). A frequently emitted one is DECSCUSR (CSI Ps SP q), used by applications that select a cursor shape per editing mode.

Changes

Add EScsiInter, entered from EScsi on an intermediate byte. It consumes any further intermediate bytes and then the final byte, so nothing reaches the screen. Sequences arriving here are not implemented by fbterm and are discarded silently.

This also keeps CSI 2 SP q (DECSCUSR) distinct from CSI 2 q (DECLL), which is implemented as set_led — the two are different control functions and must not be conflated.

The existing entry for '%' in EScsi, carrying the comment "otherwise the trailing 'm' gets printed", addressed the same problem for a single byte. It is moved into the new state, where the trailing m reaches the SGR handler through EScsiInter. ECMA-48 assigns no function to an intermediate byte followed by 06/13, so this does not shadow a standard sequence.

ESkeep is used as NR_STATES and terminates the section walk in init_state(), so the new state is inserted before it; a comment now records that constraint.

Section numbers in the escape_sequences[] comments were off by one from ESgreater onwards; corrected while adding the new section.

Testing

printf 'expect AB   : A\033[2 qB\n'
printf 'expect RED  : \033[31mRED\033[0%%m NORMAL\n'
printf 'expect CD   : C\033[2qD\n'

	before	after
A CSI 2 SP q B	AqB	AB
CSI 0 % m	resets SGR	resets SGR
C CSI 2 q D (DECLL)	CD	CD

The second line checks that moving the '%' entry did not regress it; the third checks that DECSCUSR and DECLL remain distinct.

Built and run on x86_64 (Arch, Linux 6.16) and armv7l (Debian bookworm, Linux 3.10), reading /dev/fb0 before and after on both. Colour output, UTF-8 input via uim-fep and scrollback are unaffected.