po 13. 7. 2020 v 16:57 odesílatel Daniel Gustafsson <
daniel@yesql.se> napsal:
> On 13 Jul 2020, at 13:02, Pavel Stehule <pavel.stehule@gmail.com> wrote:
> I like JSON format. But why here? For this purpose the JSON is over engineered.
I respectfully disagree, JSON is a commonly used and known format in systems
administration and most importantly: we already have code to parse it in the
frontend.
I disagree with the idea so if we have a client side JSON parser we have to use it everywhere.
For this case, parsing JSON means more code, not less. I checked the parse_manifest.c. More
the JSON API is DOM type. For this purpose the SAX type is better. But still, things should be simple as possible.
There is not any necessity to use it.
JSON is good for a lot of purposes, and can be good if the document uses more lexer types, numeric, ... But nothing is used there
> This input file has no nested structure - it is just a stream of lines.
Well, it has a set of object types which in turn have objects. There is more
structure than meets the eye.
Also, the current patch allows arbitrary whitespace before object names, but no
whitespace before comments etc. Using something where the rules of parsing are
known is rarely a bad thing.
if I know - JSON hasn't comments at all.
> I don't think so introducing JSON here can be a good idea.
Quite possibly it isn't, but not discussing options seems like a worse idea so
I wanted to bring it up.
> It is a really different case than pg_dump manifest file - in this case, in this case pg_dump is consument.
Right, as I said these are two different, while tangentially related, things.
Backup manifest format has no trivial complexity - and using JSON has sense. Input filter file is a trivial - +/- list of strings (and it will be everytime).
In this case I don't see any benefits from JSON - on both sides (producent, consuments). It is harder (little bit) to parse it, it is harder (little bit) to generate it.
Regards
Pavel
cheers ./daniel