[qmtest] TET result stream

Mark Mitchell mark at codesourcery.com
Fri Feb 20 09:31:28 UTC 2004


Nathaniel Smith wrote:

>On Thu, Feb 19, 2004 at 12:01:44PM -0800, Mark Mitchell wrote:
>  
>
>>Nathaniel Smith wrote:
>>    
>>
>>>                       "%i %s 00:00:00"
>>>      
>>>
>>As I mentioned on IRC, I'd like to see us use a valid time here.  I 
>>suggest extending QMTest to collect start/stop time for tests; I can see 
>>that being useful to other people as well.
>>    
>>
>
>Hmm, okay.  The thought had occurred to me, but I figured you wouldn't
>like the size increase in results files -- a bit over 40 bytes per
>result -- based on some conversations we had a while ago when I was
>hacking on the pickle format.  This adds more functionality, of
>course...
>  
>
Yes.  I'm starting to conclude that either (a) users should not mind 
having giant results files, or (b) they should use something like the 
SQL interface you worked on to store the data on a big old pile of disks 
somewhere, or (c) not run too many tests. :-)

>Patch attached for reviews.
>  
>
OK, check it in.

>Question: current patch assumes that dejagnu subtests all start and
>end simultaneously with their containing test.  I know this is true
>for the compiler tests, where subtests are basically different ways
>of scanning compiler output, but if it's not true in general than I'd
>rather lie obviously then lie subtly... (i.e., switch back to using
>00:00:00 for purpose start/end times, so it's obvious that the times
>can't be trusted).
>  
>
I'm afraid that something like 00:00:00 will (a) be coincidentally right 
twice a day, and (b) result in the appearance that time is going 
backwards for some tool, with corresponding weird behavior.  The subtle 
lie is at least logically consistent.

I'd not sweat this detail too much -- we can cross this bridge when we 
have tons of users running old non-compiler DejaGNU tests with the Tet 
output format.  Note that I do say "when", not "if", because it is 
obvious that this will in fact become a dominant use case for QMTest. :-)

-- 
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark at codesourcery.com




More information about the qmtest mailing list