(resend) msi: Fix bug in handling of multivolume CAB files.

[Ove Kaaven]

James Hawkins skrev:
> On Wed, Mar 5, 2008 at 9:15 PM, Ove Kaaven <ovek at transgaming.com> wrote:
>> James Hawkins skrev:
>>
>>> The tests that now pass with native cabinet dll are test_continuouscab
>>  > (which is similar to what you're trying to test).  The point of
>>  > maxsize is so that it creates continuous cabs...there's no other way
>>  > to do it, and builtin doesn't create continuous cabs at all.
>>
>>  No? Then I really wonder what those hundreds of .cab files it creates
>>  from that 50k testfile is, if not continuous cabs. What else are they?
>>
> 
> Sorry, I meant that it doesn't create the compressed continuous cabs
> that the tests expect.

Well, that's because create_cc_test_files expect to get 2 cab files, not 
600, and so the "caesar" part proceeds to overwrite the 3rd in the 
series, rendering the whole series unreadable. By adjusting the max size 
so that only 2 (uncompressed) cab files are created, most of the todo 
tests appear to run fine. Just changing the file name might also work. 
(Actually I don't see any reason your tests would really need to depend 
on compression anyway, it's supposed to be a test of MSI, not cabinet.)

>>  Here's what I managed to put together that seems to trigger the problem
>>  condition... the cab extraction sequence was somehow different in my
>>  real-world case, so I had to fudge this testcase a little by packing a
>>  third file into the continuous cab. So what do you think about this
>>  test, then?
>>
> 
> Looks good to me, as long as the test fails without your fix and
> succeeds with your fix (with the cabinet override of course).

Actually, I think it works with both builtin and native cabinet, because 
my create_cc2_test_files creates uncompressed continuous cabinets, which 
both builtin and native seem to handle. (I was almost forced to use 
uncompressed anyway, because create_file creates sparse files, which 
compresses into almost nothing, which made it difficult to predict 
whether the contents would be distributed the way I needed...)