| Memory leak in STRMID [message #92105] |
Wed, 14 October 2015 04:26  |
Heinz Stege
Messages: 189
Registered: January 2003
|
Senior Member |
|
|
Hi all,
I believe that there is a bug within IDL's strmid function. However
I'm using a 5 years old version of IDL. So this bug may be fixed in
the meantime.
There seems to be a memory leak in the strmid funcion. When you
calculate a substring c of a very long string b
c=strmid(b,1000,10)
then the short string c (10 chars in this case) needs as much memory
as the long string b.
Here is the demo (and a workaround):
First I start a fresh IDL session an look for the memory consumption:
|IDL> help,/mem
|heap memory used: 797011, max: 800529, gets: 1137, frees: 291
About 1 MB overhead. Seems to be okay for running the workbench. Then
I create a huge string with 10 million chars:
|IDL> b=byte((randomu(seed,10000000,/long) mod 255)+1)
|IDL> help,/mem
|heap memory used: 10797258, max: 50797356, gets: 1152, frees: 304
|IDL> b=string(b)
|IDL> help,/mem
|heap memory used: 10797072, max: 20797211, gets: 1164, frees: 316
The string needs 10 MB memory. So far so good. Now I create a
substring with strmid:
|IDL> c=strmid(b,1000,10)
|IDL> help,/mem
|heap memory used: 20796961, max: 20797050, gets: 1176, frees: 327
Oops, the result string with 10 chars needs 10 MB too? Is it really 10
chars long?
|IDL> help,strlen(b)
|<Expression> LONG = 10000000
|IDL> help,strlen(c)
|<Expression> LONG = 10
|IDL> help,/mem
|heap memory used: 20796822, max: 20797026, gets: 1197, frees: 348
Yes it is. 10 MB for 10 chars! Can we make a copy of c, and how much
memory will it need?
|IDL> cc=c
|IDL> help,/mem
|heap memory used: 20796753, max: 20796833, gets: 1209, frees: 359
The copy of c needs only a few byts. That's fine. How can we release
the unjustified memory usage of c? Does a simple operation help?
|IDL> c+=''
|IDL> help,/mem
|heap memory used: 10796631, max: 20796759, gets: 1221, frees: 371
Aah, it helps. This is a workaround.
|IDL> print,!version
|{ x86 Win32 Windows Microsoft Windows 8.0.1 Oct 5 2010 32 64}
And now it's up to someone of you, to check verion 8.5. Unfortunately
I don't have this version.
Cheers, Heinz
|
|
|
|
|
|
|
|
| Re: Memory leak in STRMID [message #92110 is a reply to message #92109] |
Wed, 14 October 2015 10:27  |
chris_torrence@NOSPAM
Messages: 528
Registered: March 2007
|
Senior Member |
|
|
On Wednesday, October 14, 2015 at 10:15:34 AM UTC-6, Chris Torrence wrote:
> On Wednesday, October 14, 2015 at 5:33:18 AM UTC-6, superchromix wrote:
>> I can confirm that the same behaviour occurs in IDL 8.5.
>>
>> nice catch
>>
>> Mark
>
> Hmmm, well, here's the comment I found in the strmid function:
>
> * NOTE: This routine does not re-allocate memory to hold the
> * shortened string in order to avoid unneccessary fragmentation
> * of virtual memory. It simply uses the existing string memory
> * (moving the trailing null character up) unless the resulting string is
> * null (in which case it is freed).
>
> So, bug? Feature? Undesirable side effect?
>
> -Chris
Okay, I fixed this.
Just FYI, this wasn't a "leak" - IDL was still keeping track of the memory, and doing a .reset would free up the memory.
STRMID now creates new strings of the correct length, regardless of whether the input strings are temporaries or not. This may lead to some memory fragmentation, but that's probably better than wasting a bunch of memory.
Thanks for reporting it!
-Chris
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
