fix: prevent trailing backslash from "eating" triple quotes

[thalesmg]

Fixes https://emqx.atlassian.net/browse/EMQX-14157

[zmstone]

could you help to explain why this "eat" happens?

This should work?

([^"]|"[^"]|""[^"]|\\{EscapeNoQuote}|\\{UnicodeEscape})

[thalesmg]

The eat happens because the dangling backslash is the last thing the triple quoted string. ...\\""" essentially eats the last quote and breaks out of the triple quote.

"""aaa\"""\nx = 10\ny = """~...

Since \ was not being escaped when "unparsing" the config, it would eat the triple quote, and everything after that gets interpreted as the string until it hits another triple quote.

It's easy to see if you try the test case out without the code changes.

[thalesmg]

For the record (with added print debug to hocon_scanner:unindent's 3rd clause):

======================== EUnit ========================
hocon_pp_tests: triple_quote_string_ending_in_backslash_test...
----------------------------------------------------
2025-04-22 15:10:25.849
{nonode@nohost,hocon_scanner,132,<0.388.0>}>>>>>>>>>
  #{chars => "\t\"\\\"\\t\\\"\"\"\n}\nroot2 {\n  x = "}

*failed*
in function erlymatch:run/5 (/home/thales/dev/emqx/hocon/_build/test/lib/erlymatch/src/erlymatch.erl, line 36)
in call from hocon_pp_tests:triple_quote_string_ending_in_backslash_test/0 (/home/thales/dev/emqx/hocon/test/hocon_pp_tests.erl, line 359)
in call from eunit_test:'-mf_wrapper/2-fun-0-'/2 (eunit_test.erl, line 274)
in call from eunit_test:run_testfun/1 (eunit_test.erl, line 72)
in call from eunit_proc:run_test/1 (eunit_proc.erl, line 544)
in call from eunit_proc:with_timeout/3 (eunit_proc.erl, line 369)
in call from eunit_proc:handle_test/2 (eunit_proc.erl, line 527)
in call from eunit_proc:tests_inorder/3 (eunit_proc.erl, line 469)
**throw:{mismatch,{hocon_pp_tests,359,
                          {details,tuple,mismatch,
                                   [{1,value,mismatch,{ok,error}},
                                    {2,value,mismatch,
                                     {#{<<"root1">> =>
                                            #{<<"x">> => <<"\t\"\\\""...>>},
                                        <<"root2">> =>
                                            #{<<"x">> => <<"sele"...>>}},
                                      {scan_error,#{...}}}}]}}}
  output:<<"Warning: ct_logs not started
{nonode@nohost,hocon_scanner,132,<0.388.0>}>>>>>>>>>
  #{chars => "\t\"\\\"\\t\\\"\"\"\n}\nroot2 {\n  x = "}Match failed in module 'hocon_pp_tests' at line 359:
  details: {...}
            1: EXPECT = ok
                  GOT = error
            2: EXPECT = #{<<"root1">> => #{<<"x">> => <<"\t\"\\\"\\t\\">>},<<"root2">> => #{<<"x">> => <<"select \n from\n \"hello\" ">>}}
                  GOT = {scan_error,#{line => 5,reason => "illegal characters \"~\""}}
">>

=======================================================
  Failed: 1.  Skipped: 0.  Passed: 0.
===> Error running tests

[thalesmg]

I thought I had tried your suggestion before, but paired with other changes (escaping or not escaping the string, various combinations of changes), but didn't work before, but now the added test case works with that regex. 🤔

[zmstone]

\\{Escap} is changed to \\{EscapenoQuote}
does it mean \\ is now parsed as \\, but not \ ?

[zmstone]

nvm, \\ is in EscapNoQuote, it's "no quote", not "no backslash".

I wonder if \\ should have been added in MultilineChar at all.
it's a breaking change if we remove it though.

[thalesmg]

not sure I understand the question.

the regex for EscapeNoQuote is simply the old Escape without ".

[thalesmg]

I've just tried your suggestion, and with the fix I added to escape contents of triple quotes that are not multiline, it seems to work.

[thalesmg]

P.S.: somehow, GH showed only your 1st comment when I wrote my 1st reply 🙈 🙉 🙊

[thalesmg]

tagged 0.45.3