On 09/16/2013 08:02 PM, Jiří Zárevúcky wrote:
> On 16 September 2013 16:03, Martin Decky <[email protected]
> <mailto:[email protected]>> wrote:
> 
>         I don't really care about the actual prefix. All of file_*, dir_*,
>         vfs_*, vcl_* are fine with me.
> 
> 
>     I do care, but obviously this is a matter of personal taste and the
>     discussion can be endless. Thus anyone with mainline commit access
>     has the liberty to merge anything he/she considers reasonable.
> 
>     And I have the liberty to alter the code later on if I don't
>     consider it reasonable :)
> 
> 
> How about this: The pointer handle represents a generic filesystem node
> just for the purposes of navigating the client's local image of the
> hierarchy, but *_open() returns a numerical handle that is used in
> vfs_read(), vfs_write(), vfs_truncate(), etc.
> I.e. the functions operating on file's contents would be simple IPC
> wrappers, and the fat stuff is used just for navigating the file tree.
> 
> Although, I can't figure out what would be a proper name and prefix for
> the "generic fs node" type. Both "file" and "dir" suggest a particular
> kind of file, and "node" or "tree" are both too non-obvious.

Would not this defeat the orthogonality of the API? What if you want to
use the handle returned by *_open() for something else than operations
on the content (client-based binding comes to mind). To me, the API was
more beautiful than the Unix API, because there is a single object type
that all methods work with (not really important whether a numeric
handle or an opaque struct).

Jakub

_______________________________________________
HelenOS-devel mailing list
[email protected]
http://lists.modry.cz/listinfo/helenos-devel

Reply via email to