[Vm-dev] FilePlugin>>primitiveFileStdioHandles fails incorrectly?

Fri Aug 10 11:32:31 UTC 2018

Hi Denis,

> On Aug 9, 2018, at 1:18 PM, Denis Kudriashov <dionisiydk at gmail.com> wrote:
> 
> 
> 
> 2018-08-09 17:34 GMT+01:00 Eliot Miranda <eliot.miranda at gmail.com>:
>>  
>> Hi Alistair,
>> 
>>> On Thu, Aug 9, 2018 at 1:37 AM, Alistair Grant <akgrant0710 at gmail.com> wrote:
>>>  
>>> Hi Eliot,
>>> 
>>> On Wed, 8 Aug 2018 at 21:31, Eliot Miranda <eliot.miranda at gmail.com> wrote:
>>> >
>>> >
>>> > Hi Alistair,
>>> >
>>> > On Sun, Aug 5, 2018 at 1:02 PM, Alistair Grant <akgrant0710 at gmail.com> wrote:
>>> >>
>>> >>
>>> >> Hi Everyone,
>>> >>
>>> >> FilePlugin>>primitiveFileStdioHandles is responsible for opening stdio
>>> >> in the VM.  The basic behaviour is to call sqFileStdioHandlesInto()
>>> >> which does the actual work and returns a mask indicating which streams
>>> >> were successfully opened (stdin, stdout, stderr).
>>> >>
>>> >> However primitiveFileStdioHandles regards a mask of 0, i.e. no files
>>> >> available, as a primitive failure:
>>> >>
>>> >>
>>> >> sHFAfn ~= 0 ifTrue:
>>> >>     [(self cCode: ' ((sqInt (*)(void))sHFAfn)()' inSmalltalk: [true]) ifFalse:
>>> >>         [^interpreterProxy primitiveFailFor: PrimErrUnsupported]].
>>> >> self cCode: '' inSmalltalk: [fileRecords := Array new: 3].
>>> >> validMask := self sqFileStdioHandlesInto: fileRecords.
>>> >> validMask = 0 ifTrue:
>>> >>     [^interpreterProxy primitiveFailFor: PrimErrUnsupported].
>>> >>
>>> >>
>>> >> This doesn't cause a problem in squeak since all primitive errors are
>>> >> handled by simply assuming that the stdio files can't be opened and
>>> >> ignoring any other errors.
>>> >>
>>> >> Pharo currently raises an unhandled exception.
>>> >>
>>> >> On Windows not having stdio available is a normal condition, so
>>> >> I think the correct behaviour is for the primitive to succeed, but
>>> >> return an array of 3 nils, i.e. successfully determined that none of the
>>> >> stdio files are available (this is effectively what happens in the
>>> >> Squeak image anyway).
>>> >>
>>> >> It would then mean that a primitive failure can be treated as something
>>> >> going wrong, rather than the normal (on Windows) situation of no stdio
>>> >> files being available.
>>> >>
>>> >> The other (minor) benefit is that at the moment there are two situations
>>> >> in which the primitive fails with PrimErrUnsupported: sHFAfn ~= 0 and
>>> >> validMask = 0.  This change would make it easier to interpret the
>>> >> failure.
>>> >>
>>> >> Any objections?
>>> >
>>> >
>>> > No, but I think the right thing to do is to fail if  sqFileStdioHandlesInto answers a value less than zero.  That would;d still allow the subsystem to communicate some serious error, while allowing the system also to report that no streams are available (e.g. when not attached to a console).
>>> 
>>> Wouldn't it  be better for sqFileStdioHandlesInto to flag the
>>> primitive as failing itself as it could then provide a reason, e.g.
>>> primitiveFailFor or primitiveFailForOSError? (the platform support
>>> file is already doing this in other support routines).
>>> 
>>> 
>>> > BTW, is there any image level source code available for accessing the FileAttributesPlugin's primitives?  This is lazy of me, but if there is could you email me a pointer to the code or the code itself asap?  TIA
>>> >
>>> > P.S> I'm specifically looking for code to implement testing directory writability in Squeak.
>>> 
>>> OK, just a warning in advance: The code (VM and image) works fine on
>>> Linux and Windows.  I didn't know about the Mac file encoding quirks
>>> until recently, so need to change the plugin to work properly on Mac
>>> OS.  But I haven't got the VM to compile on my Mac virtualbox machine,
>>> so the changes are still local to my PC.  This won't affect you unless
>>> you've got non-ascii characters in your file names.  I should have
>>> fixed this by now, but the problems compiling on Mac have frustrated
>>> me, and I've had very little time (personal reasons).
>> 
>> I develop on the Mac so feel free to email me a change set and tests (e.g. just doit in a workspace) and IO can test for you.
>>  
>>> 
>>> The main repository for the image side file attributes code is at:
>>> https://github.com/akgrant43/pharo/tree/21368-Integrate-FileAttributesPlugin.
>> 
>> OK, I'll see if I can grab this wen I have time. Today is taken :-)
>>  
>>> 
>>> If you're not comfortable with git I think figuring out the changes
>>> will be difficult.  I've got a .cs from a slightly earlier revision
>>> which I can email you - probably directly, it's 93Kb.
>>> 
>>> I'll try and put together a change set which just includes enough code
>>> to get file attributes.
>> 
>> I've got myself going.  I have the R_OK, W_OK, E_OK accessors in place so don't need source code.  I do suggest however that primitiveFileAttribute would be nicer to use if it took its arguments in reverse order.  i.e.
>> 
>>     attribute: attributeNumber forPath: path
>> 
>> is less clumsy than
>> 
>>     attributeForPath: path numbered: attributeNumber
> 
> It can be with more traditional #at: 
> 
>      attributeForPath: path at: attributeNumber

Let’s analyze:

    attribute: attributeNumber forPath: path
has three words in the selector:
    attribute for Path
six words in the message pattern
    attribute attribute Number for Path path
which could be five and read just as well:
attribute n for Path path

    attributeForPath: path at: attributeNumber
has four words in the selector:
    attribute For Path  at
seven words in the message pattern
    attribute For Path path at attribute Number
which could be six and read just as well
    attribute For path at attribute Number

So putting the path first forces the selector and the message pattern to be longer.  It is clumsy.

> 
>> 
>>> 
>>> Any other questions, of course please let me know.
>>> 
>>> Cheers,
>>> Alistair
>> 
>> 
>> _,,,^..^,,,_
>> best, Eliot
>> 
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20180810/ae13e42c/attachment.html>