more Ada.Directories questions

more Ada.Directories questions

[tmoran]

Does Ada.Directories.Create_Directory("X"); raise Use_Error if directory X
already exists?  It seems it should if the system does not support
creation of something that already exists.  (On Windows at least, mkdir x
fails if x already exists.) Ditto if file "X" exists and the system won't
allow a file and a directory of the same name.  Does "Create" mean "change
the state of the file system" or does it mean "arrange that the file
system has a certain state"?  Create_Directory apparently means the
former.

Apparently the latter meaning applies to Create_Path, since
according to the Rationale, Ada.Directories.Create_Path("X");
does not raise Use_Error, but is a NOP, if directory X exists.

So the only reason to use Create_Directory rather than Create_Path is
to detect the case where a file or directory of that name already
exists, correct?

If "???" is an illegal name for a directory, it seems that
Create_Path("a/b/???/c"); will raise Name_Error without creating "a/b".
Correct?

Re: more Ada.Directories questions

[Adam Beneschan]

On Mar 31, 10:49 pm, tmo...@acm.org wrote:
> Does Ada.Directories.Create_Directory("X"); raise Use_Error if directory X
> already exists?  It seems it should if the system does not support
> creation of something that already exists.  (On Windows at least, mkdir x
> fails if x already exists.) Ditto if file "X" exists and the system won't
> allow a file and a directory of the same name.  Does "Create" mean "change
> the state of the file system" or does it mean "arrange that the file
> system has a certain state"?  Create_Directory apparently means the
> former.
>
> Apparently the latter meaning applies to Create_Path, since
> according to the Rationale, Ada.Directories.Create_Path("X");
> does not raise Use_Error, but is a NOP, if directory X exists.

I wonder if this was the intent.  If you say Create_Path("/a/b/c"),
then clearly the intent is to create /a and /a/b if they don't exist,
but use the existing ones if they do.  I'm not certain that that was
supposed to apply to the final directory name, /a/b/c.


> So the only reason to use Create_Directory rather than Create_Path is
> to detect the case where a file or directory of that name already
> exists, correct?

Incorrect, I think.  Create_Path will create other directories along
the path if they don't exist, and that may not be what you want.
Suppose Arg is a command-line argument and I want to say something
like

   Create_Directory (Arg & "/Work_Directory");

[actually, I'd use other routines to create the path name instead of
"&", but that's not relevant to my point].  If Arg includes a
directory component that doesn't exist because it's misspelled, I'd
want Create_Directory to fail---not to go ahead and create a new
misspelled path.


> If "???" is an illegal name for a directory, it seems that
> Create_Path("a/b/???/c"); will raise Name_Error without creating "a/b".
> Correct?

I don't think the RM is clear on this.  In fact, it may be unfeasible
to implement Create_Path in a way that doesn't attempt to create
earlier directories before finding out that a later one has a bad
name, if the implementation depends on the OS to tell it whether a
name is bad or not.  In fact, this raises another question: suppose a/
b cannot be created for some reason (no permission), then does the
above Create_Path call raise Use_Error or Name_Error?  My guess is
"not specified by the language".

                                -- Adam

Re: more Ada.Directories questions

[Randy Brukardt]

"Adam Beneschan" <adam@irvine.com> wrote in message 
news:a2b3581f-47a5-47a1-84f9-28fb819f7284@n2g2000prj.googlegroups.com...
>On Mar 31, 10:49 pm, tmo...@acm.org wrote:
>> Does Ada.Directories.Create_Directory("X"); raise Use_Error if directory 
>> X
>> already exists? It seems it should if the system does not support
>> creation of something that already exists. (On Windows at least, mkdir x
>> fails if x already exists.) Ditto if file "X" exists and the system won't
>> allow a file and a directory of the same name. Does "Create" mean "change
>> the state of the file system" or does it mean "arrange that the file
>> system has a certain state"? Create_Directory apparently means the
>> former.
>>
>> Apparently the latter meaning applies to Create_Path, since
>> according to the Rationale, Ada.Directories.Create_Path("X");
>> does not raise Use_Error, but is a NOP, if directory X exists.
>
>I wonder if this was the intent.  If you say Create_Path("/a/b/c"),
>then clearly the intent is to create /a and /a/b if they don't exist,
>but use the existing ones if they do.  I'm not certain that that was
>supposed to apply to the final directory name, /a/b/c.

The description of Create_Path starts out "Creates zero or more 
directories...". So it seems pretty clear that it can do nothing when it is 
called.

I know that I expect it to be usable as a precaution (that is, call this 
just in case the directory doesn't exist). It sort of is an assertion that 
this directory exists; and that's how I tend to use it.

Note that since these operations intrinsicly have race conditions (with 
other things going on a machine), testing for existence and creating only if 
it fails is not certain to work, so having a single routine to do the entire 
operation is better.

>> If "???" is an illegal name for a directory, it seems that
>> Create_Path("a/b/???/c"); will raise Name_Error without creating "a/b".
>> Correct?

>I don't think the RM is clear on this.  In fact, it may be unfeasible
>to implement Create_Path in a way that doesn't attempt to create
>earlier directories before finding out that a later one has a bad
>name, if the implementation depends on the OS to tell it whether a
>name is bad or not.  In fact, this raises another question: suppose a/
>b cannot be created for some reason (no permission), then does the
>above Create_Path call raise Use_Error or Name_Error?  My guess is
>"not specified by the language".

While I doubt that there will be an ACATS test for it, the language seems 
reasonably clear that Name_Error is tested first - for the whole string. 
It's mentioned first, and it is described as a test on the whole string.

I think there would be some argument that that is over-specification, and I 
don't see any great reason to insist on it (thus no ACATS test).

I do hope that you won't turn this answer into an Ada-Comment; you already 
have introduced many of the least important AIs ever... :-)

                                 Randy.


                                -- Adam