[PATCH, libgfortran] PR 25697 FIX Segmentation fault/ Bad Address on unformatted read

[Jerry DeLisle]

Steve Kargl wrote:
> On Mon, Jan 16, 2006 at 05:37:25PM -0800, Jerry DeLisle wrote:
> 
>>Steve Kargl wrote:
>>
>>>On Sat, Jan 14, 2006 at 12:30:12PM -0800, Jerry DeLisle wrote:
>>>
>>>
>>>>BTW, still waiting approval on pr25631 (the last of the tl format bugs 
>>>>(ha ha)).
>>>>
>>>
>>>This is OK (although I can't find your original email in my inbox).
>>>
>>
>>Thanks Steve, ignore my PING on 25631 I did not see this ahead because of 
>>the subject line.
>>
>>BTW on pr25697 I have tracked it partly to differences between the small 
>>buffer (8192 bytes) and the large buffer logic of fd_alloc.  Hope to look 
>>at it some more tonight.
> 
> 
> Janne has submitted a PR about ripping out the current IO buffering
> scheme in gfortran.  What you describe above seems to be just another
> symptom that we're treating.  Perhaps, it is time to make an effort
> at changing over to using read/write or fread/fwrite directly.
> Janne, have you looked into this after you submitted the PR?
> 
I found the breakage and I am working the fix now.  I saw Janne's suggestion, 
but I am afraid if we diddle with it, it will break everything again.  I am 
biased of course because I have spent considerable time getting the I/O working 
correctly and this is the last bug that Dale Ranta has been able to turn up. 
(with his mega permutations tests).

I realize if we are careful and keep the interfaces to the routines in unix.c, 
it should transition clean (cough).

I have a philosophy:  If its not broke don't fix it.  Regardless, if I can get 
this one fixed and into 4.1, it will be good.

Jerry