[Insight-developers] [Insight-users] 50 character limit in ITK on Win32?

Thu May 3 20:03:49 EDT 2012

Here's the problem: the 50 character limit doesn't actually fix the
problem.  From Andras' links, the exact allowable space for various
recent versions of visual studio can be determined.

If the formula from Andras' first link is used, the hard path limit
for vs2005 (maybe vs2008?) can be calculated.  Since it's project
related, having projects named "ITKMarkovRandomFieldsClassifiers-all"
and "ITKGPUPDEDeformableRegistration-all" means that the hard limit is

PATH + 1 +
"ITK"(3) + 1 +
"ITKMarkovRandomFieldsClassifiers-all" (36) + 1 +
"ITKMarkovRandomFieldsClassifiers-all" (36) + 1 +
80
or
PATH + 159 characters.

This formula may only work for vs2005, but it pretty clearly shows
that for vs2005, 50 characters seems a bit conservative.  I assume
that 2008 uses a similar formula, since 66 characters for path length
appears to be fine.  Since I've never really had a problem with ITK
compilation before using 2008, I guess I never had paths in the 100
character range before.

In 2010, according to Andras' links, the problem is relative pathing.
When project names are allowed to grow, and there are a large number
of nested directories where various vcproj files can be located, then
the limit will need to be more conservative.  It's 51 now, but as soon
as someone decides to make
"ITKMyAwesomeProjectUsingGradientsAndCurvatureMeasurements-all" or
some other similarly long name, vs2010 (and perhaps vs2005 and vs2008)
compilation just won't work.  Nightly dashboards will (should) catch
the problem, of course, but it seems counterproductive to limit
windows users when a stylistic change will allow these users to
proceed without any issues. Does this proposed style to limit names
harken back to days of fitting everything into 8.3 characters?
Maybe-- but it would address the problem a bit more directly than a
hard limit discovered by trial and error.

Limiting project names to 20 characters would free up the worst case
to have upwards of 30 more characters in the path, if I'm reading
these formulas correctly.  It should also give vs2010 users a bit more
breathing room, especially those who run on machines where they can't
change their default path directories due to security constraints.

Just my thoughts,
Mark
On Thu, May 3, 2012 at 4:13 PM, David Cole <david.cole at kitware.com> wrote:
> On Thu, May 3, 2012 at 6:44 PM, Andras Lasso <lasso at cs.queensu.ca> wrote:
>>> For those of us with vs2008, it's just fine to have a 66 character (or
>> longer) path
>>
>> This is not always true. The maximum allowable path length is also dependent
>> on length of the solution name, project names, and dependency between
>> projects. See:
>> http://social.msdn.microsoft.com/forums/en-US/csharpide/thread/d614d0ba-eea2
>> -444b-9be8-7fe4fb85a226/
>> http://support.microsoft.com/kb/2516078
>>
>> Andras
>>
>
> Thanks for the link. Another piece of evidence that reinforces my
> "opinion." I will continue to strongly, strongly recommend keeping
> file and directory names short for those that need to build on
> Windows. :-)
>
>
>
>> -----Original Message-----
>> From: insight-developers-bounces at itk.org
>> [mailto:insight-developers-bounces at itk.org] On Behalf Of Mark Roden
>> Sent: Thursday, May 03, 2012 6:18 PM
>> To: David Cole
>> Cc: Insight Developers
>> Subject: Re: [Insight-developers] [Insight-users] 50 character limit in ITK
>> on Win32?
>>
>> Hi David,
>>
>> I am locally removing the check for now.  My concern was that I'm developing
>> a deployment script where, rather than having to check binaries into source
>> control, the script will sync to the itk repo, to a particular commit, and
>> then build that commit, starting from scratch and checking the appropriate
>> cmake options.  I just want to avoid having to remove the check in the
>> script, since it seems superfluous to my build process anyway.
>>
>> I'd also suggest that there be a maximum character file name style
>> convention along with all of the other style requirements that ITK has.  I
>> don't think Windows' 255 character path limit is going away any time soon,
>> and in light of that, using file names like
>> ConstNeighborhoodIteratorGetBoundaryConditionChange.xml
>> and
>> itkLevelSetEvolutionNumberOfIterationsStoppingCriterionTest.cxx
>> would invite this kind of problem, especially if the files are buried deep
>> into several nested directories.
>>
>> I'll look out for CMake 2.8.9 and update to it once it's out.
>>
>> Thanks,
>> Mark
>>
>>
>> On Thu, May 3, 2012 at 3:04 PM, David Cole <david.cole at kitware.com> wrote:
>>> On Thu, May 3, 2012 at 5:30 PM, Mark Roden <mmroden at gmail.com> wrote:
>>>> Hi Cory,
>>>>
>>>> OK, I'll switch over to the developer's thread.
>>>>
>>>> But it appears that the 50 character limit (or 51 character limit, as
>>>> reported by Alex Gouaillard) is only a problem with a particular
>>>> combination of vs2010 and cmake 2.8.3.  For those of us with vs2008,
>>>> it's just fine to have a 66 character (or longer) path.
>>>>
>>>> I would propose, then, that this error message not be for any and all
>>>> versions of win32, but only for those compiler/cmake combinations
>>>> where this bug exists.  Otherwise, this limit is just overly
>>>> restrictive.
>>>>
>>>> Thanks,
>>>> Mark
>>>>
>>>> On Thu, May 3, 2012 at 2:24 PM, Cory Quammen <cquammen at cs.unc.edu> wrote:
>>>>> Mark,
>>>>>
>>>>> I think the path length choice was made conservatively because no
>>>>> one had the patience to figure out the exact length where things go
>> wrong.
>>>>>
>>>>> The depth of the source tree, especially with the third-party
>>>>> libraries, is more of a problem that are the file name lengths in ITK.
>>>>>
>>>>> It might be more reasonable to issue a warning rather than halt with
>>>>> an error. You may want to propose this to the developers list to see
>>>>> what others think.
>>>>>
>>>>> - Cory
>>>>>
>>>>> On Thu, May 3, 2012 at 5:13 PM, Mark Roden <mmroden at gmail.com> wrote:
>>>>>> But a path length of 66 works.
>>>>>>
>>>>>> It appears that this patch was for a particular compiler (vs2010)
>>>>>> and a particular version of cmake (2.8.3).  That's a much more
>>>>>> specific issue than all win32 platforms; I'm on vs2008, and as I
>>>>>> said, 66 characters works just fine.
>>>>>>
>>>>>> And I would be strongly in favor of making this a warning, rather
>>>>>> than hard limit, if it's difficult/impossible to detect that
>>>>>> particular combination of compiler/cmake versions.  I don't, and
>>>>>> never have had, an issue building in ITK in any random directory,
>>>>>> regardless of path
>>>>>> length-- maybe because I tend to keep my path lengths lower than
>>>>>> 100 characters.
>>>>>>
>>>>>> But if this does continue to be an issue, is there some kind of
>>>>>> style guide for path length?  I see some really absurdly long
>>>>>> filenames in there, and if it really is an issue with vs2010, then
>>>>>> maybe that style should be enforced (along with all of the other commit
>> rules).
>>>>>>
>>>>>> On Thu, May 3, 2012 at 2:04 PM, Cory Quammen <cquammen at cs.unc.edu>
>> wrote:
>>>>>>> Mark,
>>>>>>>
>>>>>>> The limit is there to save your sanity :-)
>>>>>>>
>>>>>>> Did you try to compile with a path length of 100 characters? When
>>>>>>> I did, I ran into a file-not-found complaint by the compiler when
>>>>>>> compiling GDCM. Then I checked to see that the file it is
>>>>>>> complaining about does indeed exist.
>>>>>>>
>>>>>>> See the long email thread that led to this restriction here:
>>>>>>>
>>>>>>> http://www.itk.org/pipermail/insight-developers/2011-June/019268.h
>>>>>>> tml
>>>>>>>
>>>>>>> Basically, the relative paths generated by CMake and limitations
>>>>>>> in the length of paths in Visual Studio lead to this limit.
>>>>>>>
>>>>>>> Cory
>>>>>>>
>>>>>>> On Thu, May 3, 2012 at 4:49 PM, Mark Roden <mmroden at gmail.com> wrote:
>>>>>>>> Hi all,
>>>>>>>>
>>>>>>>> I've just recompiled itk 4.0, and found that there's a 50
>>>>>>>> character path limit for win32.
>>>>>>>>
>>>>>>>> What's the logic behind this?  I comment out that line, and
>>>>>>>> everything works perfectly fine.
>>>>>>>>
>>>>>>>> The filename limit for windows is 255 (260 including the drive
>>>>>>>> letter and trailing null), but each component of the path can be
>>>>>>>> 255 characters long.  MacOS has the same limit for HFS+, where
>>>>>>>> each element is locked to 255, but the total path has to be 1024 or
>> lower.
>>>>>>>> (source:
>>>>>>>> http://en.wikipedia.org/wiki/Comparison_of_file_systems).  If
>>>>>>>> you're not using windows-specific programming, but just treating the
>> path as a string, there's a 255 character limit.
>>>>>>>>
>>>>>>>> With my base file path that's 66 characters long, the longest
>>>>>>>> filename I can find is 146:
>>>>>>>>
>>>>>>>> C:\Users\mmroden\Documents\src\MyProjectsinQT\ExternalLibs\itk\IT
>>>>>>>> K\Documentation\Migration\ConstNeighborhoodIteratorGetBoundaryCon
>>>>>>>> ditionChange.xml
>>>>>>>>
>>>>>>>> That's a far cry from 255.
>>>>>>>>
>>>>>>>> Just how deep are the paths in ITK anyway?  Is the directory tree
>>>>>>>> really 205 characters long?  Or is this just someone guessing
>>>>>>>> without actually checking?
>>>>>>>>
>>>>>>>> The problem with this is that I want to incorporate ITK building
>>>>>>>> as part of a build script, rather than having to checkin binary
>>>>>>>> images into a repo.  I'd have to add a portion to the script to
>>>>>>>> remove the limit.
>>>>>>>>
>>>>>>>> What if it were just set to 100?
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> Mark
>>>>>>>> _____________________________________
>>>>>>>> Powered by www.kitware.com
>>>>>>>>
>>>>>>>> Visit other Kitware open-source projects at
>>>>>>>> http://www.kitware.com/opensource/opensource.html
>>>>>>>>
>>>>>>>> Kitware offers ITK Training Courses, for more information visit:
>>>>>>>> http://www.kitware.com/products/protraining.php
>>>>>>>>
>>>>>>>> Please keep messages on-topic and check the ITK FAQ at:
>>>>>>>> http://www.itk.org/Wiki/ITK_FAQ
>>>>>>>>
>>>>>>>> Follow this link to subscribe/unsubscribe:
>>>>>>>> http://www.itk.org/mailman/listinfo/insight-users
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Cory Quammen
>>>>>>> Research Associate
>>>>>>> Department of Computer Science
>>>>>>> The University of North Carolina at Chapel Hill
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Cory Quammen
>>>>> Research Associate
>>>>> Department of Computer Science
>>>>> The University of North Carolina at Chapel Hill
>>>> _______________________________________________
>>>> Powered by www.kitware.com
>>>>
>>>> Visit other Kitware open-source projects at
>>>> http://www.kitware.com/opensource/opensource.html
>>>>
>>>> Kitware offers ITK Training Courses, for more information visit:
>>>> http://kitware.com/products/protraining.php
>>>>
>>>> Please keep messages on-topic and check the ITK FAQ at:
>>>> http://www.itk.org/Wiki/ITK_FAQ
>>>>
>>>> Follow this link to subscribe/unsubscribe:
>>>> http://www.itk.org/mailman/listinfo/insight-developers
>>>
>>>
>>> This commit that Brad King made recently in CMake (and it will appear
>>> in the 2.8.9 release in a couple months) does exactly that: it warns
>>> when it encounters a path that's too long, and only on Visual Studio
>>> 10.
>>>
>>>
>>> http://cmake.org/gitweb?p=cmake.git;a=commitdiff;h=d931ce9fe0c26cfddb7
>>> 66901cfd1ccbc9def71c7
>>>
>>> I'd strongly, strongly encourage you to keep your Windows root path
>>> for ITK source and build trees to be as short as possible anyhow.
>>>
>>> So, after CMake 2.8.9 is required (or at least detected) we can modify
>>> the ITK CMakeLists to eliminate / conditionalize that chunk.
>>>
>>> We could turn it into a warning.... but would people pay attention and
>>> move things when they hit it, or would they plow through, and then hit
>>> a much later build error and then be mad that they wasted their time?
>>>
>>> You can certainly make it a warning, make it conditional on VS10, or
>>> just eliminate it locally.
>>>
>>>
>>> David
>> _______________________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at
>> http://www.kitware.com/opensource/opensource.html
>>
>> Kitware offers ITK Training Courses, for more information visit:
>> http://kitware.com/products/protraining.php
>>
>> Please keep messages on-topic and check the ITK FAQ at:
>> http://www.itk.org/Wiki/ITK_FAQ
>>
>> Follow this link to subscribe/unsubscribe:
>> http://www.itk.org/mailman/listinfo/insight-developers
>>
>>