unzip headersize bug fixed

[Denys Vlasenko]

On Tuesday 06 November 2007 19:30, Paul Fox wrote:
> hi --
> 
> i commited a fix for unzip yesterday, and it occurs to me that i
> should probably mention it, since it's probably a candidate for
> 1.8.1.  (the svn rev is r20369.)

It is there:
http://busybox.net/downloads/fixes-1.8.0/busybox-1.8.0-unzip.patch

> it turs out that at least some gcc versions (our is 3.4.1 for
> arm) don't fully pack structures properly -- in the case of the
> zip header, the elements of the structure are packed, but the end
> of the struct is padded from 26 out to 28 bytes.  unzip was
> changed recently to do two things:  a) verify that the structure
> is exactly 26 bytes, and b) use sizeof(zip_header) when reading
> the header.  i fixed unzip.c to a) change the size check so that
> a struct that is only too long will pass the test, and b) changed
> the read back to using a constant value of 26.
> 
> vda -- regarding your svn 20370/20372 -- are you sure that that warning
> about ATTRIBUTE_PACKED is spurious?  does ATTRIBUTE_PACKED make
> sense on a union?  i tried the same thing and backed it out
> because of the warning.

Well, I expected that PACKED there will let gcc know that entire union
should be packed too. There should be a way to suppress that extra
two bytes on ARM, right? Maybe PACKED will even work on ARM,
I didn't check, but on i386 gcc complains.
--
vda