race in verbatim extension

[saraedum]

The verbatim extension registers the character \ as a special inline character with cmark-gfm so that we can detect the string \verbatim in a comment.

However, this breaks cmark-gfm as \ is added to the (global) SPECIAL_CHARS table when inlines are processed. When the processing is complete, it is removed from that table again. This seems wrong in several ways:

• Having SPECIAL_CHARS in a global table means that cmark-gfm cannot be used in a multi-threaded environment it appears. (Fixed in Made cmark re-entrant by storing special chars data per parser object github/cmark-gfm#176)

• Also single-threaded, there are problems. Removing \ from SPECIAL_CHARS seems to be wrong, since it is a special character anyway (the escape character). Parsing a comment without the verbatim-extension now breaks detection of escapes, e.g., \`code\` understands the first escape correctly (since parse_inline() does not use the SPECIAL_CHARS table) but misses the second escape. (This is now cmark_syntax_extension_set_special_inline_chars removes standard special chars github/cmark-gfm#221.)

Btw., the way parsing works in cmark-gfm means that \verbatim cannot work whenever parse_inline() is called, e.g., at the beginning of a block since extensions are only checked after all the standard control characters have been handled.

[saraedum]

Without changing cmark-gfm, there's not much that standardese can do about this.

It's probably best to replace the \verbatim syntax with something that can be safely parsed such as `verbatim:…`. This won't work inside backticks but inside backticks we probably do not need verbatim anyway.