Enhancement: csv-parser multichar delimiter about syslog-ng

[algernon]

I suppose delimiters(string(":\n")) could be one way to specify multichar delimiters, without breaking backwards compatibility, and still making sense.

[faxm0dem]

Looks sane to me yes

[ihrwein]

What do you think about flags(string-delimiter)? It wouldn't introduce an other member variable in LogCSVParser and the code would be more consistent.

[juhaszviktor]

👍 for string-delimiter flag

[algernon]

I'm not a great fan of flags, to be honest. It means I have to keep an eye on two settings, and makes it harder to write blocks or other kinds of DSL.

[ihrwein]

A third option won: string-delimiters(":,", "//"). You can use this and delimiters() simultaneously, the former has the precedence.

I have a little ugly, but working version here: https://github.com/ihrwein/syslog-ng/tree/f/csv-string-delimeter
According to the test cases it does it's job, but I'm working on refactoring the whole code.

[algernon]

Well, that has the same problem as the string-delimiter flag: making it harder to write anything on top of it. But since I won't be using the functionality, it doesn't matter for me. Just wanted to highlight that using a new option instead of the flag doesn't change much from a DSL point of view.

[lbudai]

@algernon: thank you for your notes (Having more opinions before implementing a feature is a good thing, IMHO. )

As I see your idea, the optimal solution would be if user could set delimiters as:
delimiters(chars(",;:"), string("xx")) . But -as you already mentioned- we have to deal with backward compatibility. So backward compatibility breaks the optimal solution, in that situation, from my point of view, both solution is fine, but I would go with string-delimiters() option :-) .

@faxm0dem: which solution do yo prefer? (our opinions -as developers- not as important as yours :-) )
a)
delimiters(":,", string("//"))

b)
delimiters(":,")
string-delimiters("//")

[faxm0dem]

both would work for me, whatever's easiest to implement/maintain

[lbudai]

Ok, we prefer solution b) .

Thanks!

[algernon]

My optimal solution would be delimiters(",;:") working as before, internally perhaps being an alternative to delimiters(chars(",;:")). And only delimiters(string(",;:")) would be new.

The grammar rules for this aren't all that bad:

| KW_DELIMITERS '(' delimit-spec ')' { YYACCEPT; }

and then:

delimit-spec
 : KW_CHARS '(' string ')' { ... }
 | KW_STRING '(' string ')' { ... }
 | string { /* same code as KW_CHARS case */ }
 ;

Voila. All syntax within delimiters(), backward compatibility still preserved.

[lbudai]

@algernon: finally, your suggestion was chosen :-)

[ihrwein]

@faxm0dem, could you check it too?

[ihrwein]

The PR was merged, so this issue can be closed.