[resubmit] gdb.base, r*.exp thru w*.exp
Michael Snyder
msnyder@vmware.com
Thu May 20 21:57:00 GMT 2010
Jan Kratochvil wrote:
> On Thu, 20 May 2010 23:37:49 +0200, Michael Snyder wrote:
>> Jan Kratochvil wrote:
>>> On Thu, 20 May 2010 22:29:47 +0200, Michael Snyder wrote:
>>>> + -re ".*in main after func1.*$gdb_prompt $" {
>>> Sorry for not a complete review but for this case there should be:
>>>
>>> -re ".*in main after func1.*\r\n$gdb_prompt $" {
>>>
>>> or lib/gdb.exp proc gdb_test is using:
>>>
>>> -re ".*in main after func1.*[\r\n]+$gdb_prompt $" {
>>>
>>> as just "$gdb_prompt $" is needlessly weak in practical cases causing false
>>> positives.
>> Thanks for the review -- but I don't follow you.
>>
>> What is it about this particular case that you think requires
>> disambiguating? Or are you saying this in general?
>
> It was said in general.
>
> There could be for example output
>
> (gdb) return
> Make func1 return now? (y or n) y
> #0 main () at ./gdb.base/return.c:31
> 31 printf("in debugger (gdb) after func1\n");
> (gdb)
>
> and if expect would read(2) just the part
>
> (gdb) return
> Make func1 return now? (y or n) y
> #0 main () at ./gdb.base/return.c:31
> 31 printf("in debugger (gdb)
>
> before the next part of input arrives it can have a false termination of
> waiting on the prompt.
>
> That can happens for example for "info set" where the output contains text:
> prompt: Gdb's prompt is "(gdb) ".
>
> (In this specific testcase there is no "(gdb) " contained in the output so it
> cannot have a false positive. But examining all the testcases would be
> difficult. Also I understand even "\r\n(gdb) " can have a false positive on
> some random output.)
Well, this is a very common and ancient idiom that is used
everywhere throughout the test suite.
You have to assume that we don't care about anything between
the ".*" and the "$gdb_prompt $". If we do, the test is wrong,
but if it consumes more output than it was meant to, the
following tests will fail.
More information about the Gdb-patches
mailing list