Hi folks, 

I didn’t really know where to address this. Apologies in advance if this is the 
wrong place, it feels as good a place as any to start. 

In my day job, we recently experienced a bug that was caused by picking up an 
update to the Apache http client that includes the behaviour of automatically 
gunzipping response bodies received with a Content-Encoding: gzip response 
header. 

The problem was, we already had a client wrapper to do this further up our 
stack. Gunzip functions are not, as it turns out, a fan of already unzipped 
data :). 

We fixed the problem but it lead to an interesting debate. Should the 
http-client *remove* the Content-Encoding header when it mutates the body by 
unzipping it? Our problem was that the filter sees this header, expects the 
body to be gzipped data, but it is not. There are byte prefixes in the data for 
gzip so we can check but it does feel that the header is now inconsistent. If a 
gzipping filter would gzip the body and add the header, should not the 
decompression routine do the full inverse? 

The counter of course is that clients may need to see the full set of original 
response headers. 

I didn’t want to go charging in creating issues and things because I am aware 
that there may well be a long standing philosophy on this stuff, or some 
tremendously important thing that I am missing, so it seemed worth asking here 
first. 

Did I do a good thing?

Thanks

Wes
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to