Fix double-free on ternary operator parsing error

[pmatilai]

v1 gets double-freed when there's an error parsing the result side
of the ternary operator. Add testcases to go, one that catches the
actual error and another which serves to demonstrate that the false
branch doesn't get parsed at all, errors or no.

[mlschroe]

good catch, thanks!

[mlschroe]

The false branch does get parsed, this needs to be fixed as well.

[pmatilai]

Oh, right, got mixed up trying various different things. With eg bare words (which is where I spotted this) both the true and false clauses report the error, but with unexpected end of expression the error message is missing (but it returns with exit code 1). I'll fix the test + commit message...

[mlschroe]

I think we just need a

exprErr(state, _("syntax error in expression"), state->p);

in doPrimary()'s default case to fix missing error for %{expr:0 < 1 ? 4+ : 0}.

[pmatilai]

That'd work, but then we have inconsistent error messages depending on which branch the error is in:

$ ./rpm --eval "%{expr:0 < 1 ? 0 : 4+}"
error: unexpected end of expression: 0 < 1 ? 0 : 4+
$ ./rpm --eval "%{expr:0 < 1 ? 4+ : 0}"
error: syntax error in expression: 0 < 1 ? 4+ : 0
error:                                        ^

It also causes a duplicate error at least in the bareword case:

$ ./rpm --eval "%{expr:bare}"
error: bare words are no longer supported, please use "...": bare
error:                                                       ^
error: syntax error in expression: bare
error:                            ^

Adding TOK_TERNARY_ALT as a fallthrough case to TOK_EOF in doPrimary() seems to work but didn't actually think about it so dunno if there would be other side-effects.

[mlschroe]

The duplicate error is because rpmExprStr/rpmExprBool ignore the return value of the
rdToken() call. That should also be fixed.

It's not really about TOK_TERNARY_ALT, you'll get the same error with %{expr:4 * +}

If the different error message bothers you, just delete the TOK_EOF case so that you'll
also get a syntax error if the expression ends unexpectedly.

[pmatilai]

Added the rdToken() checks + the generic syntax error. Inconsistent error is much better than no error message at all...