Alistair did I miss some pending PR to be integrated for FS? I'm getting back to reality after a totally fun and intense ESUG conference. Stef On Thu, Sep 7, 2017 at 8:26 AM, Alistair Grant <akgrant0710@gmail.com> wrote:
Hi Andreas,
On Wed, Sep 06, 2017 at 04:53:27PM -0700, Andreas Sunardi wrote:
Hi Alistair,
I found fogbugz #19717 where this was discussed.
https://pharo.fogbugz.com/f/cases/19717/ FileSystem-workingDirectory-wrong-after-image-moved-to-a-new-folder
The issue seems to be the current working directory is saved in instance variable workingDirectory. When the image is moved to another directory, this workingDirectory becomes incorrect. Instance variable 'store', through its # defaultWorkingDirectory, always gives the right answer (however, this answer is the image directory, not directory where the command is invoked, which is the issue in #05723 you mentioned).
Here's a snippet from fogbugz #19717 original problem on how to reproduce it: QUOTE 1/ open an image. evaluate './pharo-local' asFileReference and keep the inspector. 2/ save and quit 3/ move the image to another directory 4/ open the image 5/ evaluate self fullName on the file reference => it will give a wrong path with a reference to the working directory used at file reference creation. ENDQUOTE
Thanks for finding this!
In #20164 I reintroduced caching the working directory, but update it when the session is stopped and started, so the above test produces the expected results. As you can see in the comments for the associated PR, caching the value was a significant peformance improvement for at least one person. https://pharo.fogbugz.com/f/cases/20164/Caching-DiskStore-defaultWorkingDire...
I don't know if it's still debatable, which './pharo-local' the file reference should refer to, but I think I agree with fogbugz #19717 that it should stay as relative path.
But this is a separate issue than losing the ability to change working directory, which seems to me an unintentional side-effect in the solution. I'll see if I can jump onto $05723 to request support for changing working directory.
On a side note, is current working directory the same as image directory in Pharo? I wonder why we have these issues if they are separate things. I haven't dived into this, so I can't say much.
Yes, which was the original driver for #05723. After it is integrated the working directory will be the process working directory. The image directory can of course still be accessed with FileLocator imageDirectory.
Cheers, Alistair
-- Andreas
On Wed, Sep 6, 2017 at 12:42 PM, Alistair Grant <akgrant0710@gmail.com> wrote:
On Wed, Sep 06, 2017 at 10:50:05AM -0700, Andreas Sunardi wrote: > It isn't only #changeDirectory: method that is missing. Method # > workingDirectoryPath: and instance variable workingDirectory are also missing. > > FileSystem >> changeDirectory: aPath > self workingDirectoryPath: (self resolve: aPath) > > FileSystem >> workingDirectoryPath: aPath > aPath isAbsolute > ifFalse: [ self error: 'Cannot set the working directory to a relative > path' ]. > workingDirectory := aPath > > > To solve my problem, I implemented those and also fixed # initializeWithStore: > and changed #workingDirectoryPath > > FileSystem >> initializeWithStore: aStore > store := aStore. > workingDirectory := store defaultWorkingDirectory > > FileSystem >> workingDirectoryPath > ^ workingDirectory > > > Those are from Pharo 5. However, this breaks Monticello (when opening > Monticello Browser). A FileSystem instance has instance variable > workingDirectory set to nil. This should be impossible. I set 'self halt' in # > initializeWithStore:. It doesn't get triggered, which tells me the instance > isn't created by normal way. I don't know what's going on there. > > > To fix it, I changed #workingDirectoryPath to > > FileSystem >> workingDirectoryPath > workingDirectory ifNil: [ > workingDirectory := store defaultWorkingDirectory ]. > ^ workingDirectory > > > That solves my problem, but this is a specific tool. I don't know what other > problems those changes will cause. This #ifNil: guard is not in Pharo 5, so > that makes me worry. In general, the change in FileSystem gives impression that > the it is intentional. But I haven't found the new Pharo 6.1 way to change > working directory, if there's any.
I couldn't find a fogbugz issue relating to this - if anyone knows the issue, please post it here as it would be good to understand the rationale for the change.
My interpretation of the changes are that the decision was made to hard code the working directory to the image directory. There are quite a few people who don't agree with this. :-)
There is already a proposed patch to change the working directory to the process working directory, see fogbugz #05723. https://pharo.fogbugz.com/f/cases/5723/Default-Working-Directory
The patch is waiting on a fix to include UFFI in the kernel.
It currently doesn't allow the working directory to be changed, but if I remember correctly, Rajula developed the code to change the working directory.
I think this approach has the advantage that it will automtically work with forked processes, e.g. using OSProcess / OSSubprocess.
It would be worthwhile adding a comment to the issue asking Rajula to add the change directory functionality to the patch.
Cheers, Alistair
> -- > Andreas > > On Tue, Sep 5, 2017 at 12:46 AM, Stephane Ducasse < stepharo.self@gmail.com> > wrote: > > Thanks for reporting. > I do not remember an action around me for this change. > Do you have the definition in Pharo 50 at hand? > > Stef > > On Wed, Aug 30, 2017 at 11:09 PM, Andreas Sunardi < a.sunardi@gmail.com> > wrote: > > I found FileSystem class has changed from Pharo 5 to Pharo 6. I've been > > using FileSystem>>changeDirectory to make my program (Pharo 5) runs in > the > > current working directory (thus able to find local files). > > > > This is now broken because #changeDirectory doesn't exist anymore. The > > change seems intentional. > > > > Question: Is there a different way in Pharo 6 do set working directory or > is > > this a bug? > > > > -- > > Andreas