fix: prevent infinite loop in cfIEEE1284NormalizeMakeModel on empty MFG/MDL (fixes #214)

[KHARSHAVARDHAN-eng]

Summary

Fixes an infinite loop in cfIEEE1284NormalizeMakeModel() when given IEEE-1284 device IDs containing empty MFG and MDL fields (such as MFG:;MDL:; and MFG:;MDL:;CMD:PostScript;).

Root Cause

1. Empty MFG Parsing: When MFG:; was encountered, makeptr and bufptr remained at buffer. The unpatched code checked while (isspace(*(bufptr - 1))) bufptr--;, reading before the buffer array start.
2. Invalid Space Insertion: if (bufptr < buffer + bufsize - 1) evaluated to true for empty MFG, inserting a space ' ' and advancing makeptr to buffer + 1.
3. Out-of-Bounds modelptr: When MDL:; was processed, bufptr was decremented back to buffer and NUL-terminated (""). However, if (!nomakemodel && makeptr != bufptr) assigned modelptr = makeptr (buffer + 1), causing modelptr to point past the string terminator.
4. Infinite Loop: Manufacturer deduplication calculated compare_len = modelptr - buffer (1) and repeatedly matched strncasecmp("", garbage, 1) == 0, looping infinitely in move_right_part().

Fix Description

1. Pointer Bounds Check: Added bufptr > buffer bounds checks before dereferencing *(bufptr - 1) in trailing whitespace removal loops (lines 807 & 825).
2. Empty MFG Space Check: Added bufptr > buffer condition at line 808 to prevent inserting a space separator when no manufacturer text was written.
3. modelptr Pointer Safety: Added a defensive bounds check at line 1164 (if (!modelptr || modelptr > buffer + strlen(buffer)) modelptr = buffer + strlen(buffer);) to guarantee modelptr never points past the string NUL-terminator.
4. Automated Unit Test: Created cupsfilters/test-ieee1284-normalize.c verifying that empty MFG and MDL inputs return safely without hanging or crashing, while preserving valid MDL:; and MFG:HP;MDL:DeskJet; inputs. Registered the test in Makefile.am under check_PROGRAMS and TESTS.

Fixes #214.

[zdohnal]

I would prefer returning if there is no manufacturer or model values - while the current MR might be shorter solution at the moment, but IMO it is better to end early if we don't have proper required input rather than trying to recover.

I have my proposal at zdohnal@60d2b49 , but without test. We can combine them, if preferred.

[tillkamppeter]

For me this is OK, as the bug gets fixed. If there are few devices which have actually one of the manufacturer or model field empty or missing, they can still be used.

[KHARSHAVARDHAN-eng]

Thanks @tillkamppeter and @zdohnal for reviewing this. I’ll leave the current implementation as-is based on the maintainer feedback, with the regression test covering the empty MFG/MDL cases and preserving the existing behavior for valid inputs.

[zdohnal]

@tillkamppeter my fear with that is that allowing such empty values can spawn issues further down in the ecosystem. While I would guess empty manufacturer and model values are not valid, but I don't know any standard to support this guess.

[KHARSHAVARDHAN-eng]

Thanks for the clarification. I’ve updated the implementation to return early when the required MFG or MDL value is empty, and added regression coverage for both empty and partial Device IDs while preserving the existing behavior for valid MFG + MDL inputs.