Matching http.body[0,1000] does not work for quite large files

[TafThorne]

We've run into a small problem with the v0.8.3 version of Savon. My service is reporting a SOAP fault, yet I'm not getting an exception raised by Savon.

I think my problem is due to this line in savon/soap/fault.rb:

  # Returns whether a SOAP fault is present.
  def present?
     @present ||= http.body[0,1000] =~ /<(.+:)?Body>(\s*)<(.+:)?Fault>/

In our case, the http.body is about 5k in size, with the Fault tag near the end, so I never get a match.

We have patched the file locally by replacing "http.body[0,1000]" with "http.body" so that all of the body is checked.

This issue/bug seems to still be present in v0.8.4

[rubiii]

it was implemented because with large responses, parsing the entire body with a regepx is damn slow. still trying to find a good solution for this.

[TafThorne]

I have closed this issue by accident haven't I... can someone reopen it please or shall I recreate it?

[rubiii]

i've tested quite a few possible solutions and even though i'm more comfortable with using regexps, it's 22x slower than using some native string method. maybe something rather simple like include? could work:

soap1_fault = body.include?("faultcode>") && body.include?("faultstring>")
soap2_fault = body.include?("Code>") && body.include?("Reason>")
body.include?("Fault>") && (soap1_fault || soap2_fault)

i know it's ugly, but ... opinions?

[TafThorne]

One of the guys here has made a suggestion that sounds quite good to me:
"The www.w3.org web site I was looking at is http://www.w3.org/TR/2000/NOTE-SOAP-20000508/#_Toc478383529 section 6.2
which says:

"In case of a SOAP error while processing the request, the SOAP HTTP server MUST issue an HTTP 500 "Internal Server Error" response and include a SOAP message in the response containing a SOAP Fault element (see section 4.4) indicating the SOAP processing error."

So my suggestion is that the code can perhaps search for the 500 Internal Server Error code in the HTTP response first."

This would mean that responses without a fault should be binned quite quickly based on the HTTP response and you would only have to perform the expensive and exhaustive search on files known to have a problem.

[rubiii]

thanks guys!

from my experience, relying on soap services to follow the specifications is almost always impossible. soap implementations are a mess. but sending a specific response code looks like something most services should be able to do. i'll give it a shot.

ps. could you create a gist/pastie containing your soap fault? might be interesting for regression and performance tests.

[TafThorne]

Did you want just the fault or the full response? We can probably do something about anonymising any information in the response.

[rubiii]

the complete xml (if it's not too much work anonymising private information).

[rubiii]

your problem should be fixed with v0.8.5.

[TafThorne]

Thank You.

Here is an example of our output in a gist as requested.
We are not generating the status 500 code at present but are quite happy to patch savon internally until we generate it. We have been able to get v0.8.5 to work well in our code.

[TafThorne]

I have just seen that i156 would add a configurable option, so using that until we update our server code should work well :-)

[rubiii]

thank you guys for getting involved with this issue. i really appreciate your help! your problem should be fixed with v0.8.6. i did not add a configuration option for implementations that don't follow the soap spec, but i think this would be a pretty good idea if we find any other cases like this.

thanks again!

[jkingdon]

The unconditional code, as found in 0.8.6, is fine with me. I'm not the one who had the 5 megabyte responses ;-).