Re: Experiences of XML parser generators for Ada?

[Adrien Plisson <aplisson-news@stochastique.net>]

Daniel W wrote:
> Thank you for your succinct clarification. More specifically I'm asking for 
> persons with experience of the parser generator. I actually have XMLBooster 
> downloaded, but as I said, I'm sort of short on experience.... :-)

well, i don't have any experience with parser generator (excepted with 
lex), but i would like to share my experience:

i designed a software composed of 2 parts. all parts were written in a 
different language, and each part was executing in its own context 
(think of 2 different computers). i choosed XML as the format for 
marshaled data accross the communication medium.

i first downloaded a standard XML parser (Xerces) and tried it. it was 
so slow that i could not continue with it. since i was only using a 
subset a XML (no dtd, no validation, no entity reference, only one 
encoding), i decided to write my own XML parser and XML generator. i got 
  70x performance boost.

now if i look back, i think it would have been better if i had defined 
my own protocol and not used XML:
- the xml fragment were all generated then parsed by software under my 
control, no user intervention. so there was no need for something human 
readable.
- i was mostly transmitting numeric values. since xml is a text format, 
performances were teared down by all the conversions from binary to 
string and back to binary.
- since i was mostly transmitting numeric values, all my text nodes were 
  shorter than the xml element type enclosing those values. this leads 
to HUGE overhead. e.g.: <detectionTreshold>84</detectionTreshold> 
encoded in Unicode is 84 bytes long, but the value expressed here is 
only 1 byte long.
- the only thing xml allowed me was extensibility at no cost, in a case 
were i was not really needing it.

so here comes my advice: think twice before using xml.
xml is a very powerful tool for DYNAMICALLY STRUCTURED HUMAN READABLE 
TEXT. for everything else, a basic binary protocol with some well 
defined rules to follow (endianness, size of data) will really be more 
efficient. plus, a basic binary protocol do not need complicated parsers...

here was my experience, i hope you find it useful.

-- 
rien