bsc -format changes string literals that contain line breaks, and is not idempotent on \r\r\n #9100

Found by a bug hunt over the 16 commits from eb087382 to cd641cbc.

Since #9042, formatting rewrites raw line breaks inside string literals, which changes the program. Here is a CRLF file whose string contains a raw LF:

printf 'let x = 1\r\nlet s = "a\nb"\r\n' > A.res
bsc A.res | grep 'let s'            # let s = "a\nb";
bsc -format A.res > B.res
bsc B.res | grep 'let s'            # let s = "a\r\nb";

The reverse case is an LF file whose string contains a raw CRLF (printf 'let x = 1\nlet s = "a\r\nb"\n'). Formatting turns the string "a\r\nb" into "a\nb".

Formatting is also not idempotent when a comment contains \r\r\n in an LF file:

printf 'let x = 1\n/* a\r\r\nb */\nlet y = 2\n' > C.res
bsc -format C.res > C1.res
bsc -format C1.res > C2.res
cmp C1.res C2.res                   # differ: C1 has "a\r\n", C2 has "a\n"

At b123b7c3 both cases keep the bytes unchanged.

Cause

Res_line_endings.to_lf and to_crlf rewrite the whole source text, string literals included. to_lf runs whenever the source contains a \r, and each run drops only the \r directly before a \n. They are called from res_multi_printer.ml (bsc -format), res_cli.ml, and the editor's format command (analysis/src/commands.ml).

🤖 Generated with Claude Code