Re: [PATCH] 2.4.8-pre3 NTFS update (1.1.16)

[Urban Widmark]

On Wed, 1 Aug 2001, Anton Altaparmakov wrote:

> - Simplified time conversion functions drastically without
> sacrificing accuracy, now the offending function is 3 lines instead of
> two pages of code. (-8

Does this mean I have to change the code I borrowed?
(Never mind ... :)

> -/* Converts a single wide character to a sequence of utf8 bytes.
> +/*
> + * Converts a single wide character to a sequence of utf8 bytes.
>   * The character is represented in host byte order.
> - * Returns the number of bytes, or 0 on error. */
> -static int to_utf8(ntfs_u16 c, unsigned char* buf)
> + * Returns the number of bytes, or 0 on error.
> + */
> +static int to_utf8(ntfs_u16 c, unsigned char *buf)

How is this different from utf8_wctomb in fs/nls/nls_base.c?
(in purpose)

If it is to allow ntfs_dupuni2utf8 to count the new string length, could
that be done differently? If there is a max allowed length of filenames
the "double parsing" can be avoided.

ntfs_printcb -> ntfs_encodeuni -> ntfs_dupuni2utf8 -> to_utf8

ntfs_dupuni2utf8 does a kmalloc, which is later free'd by ntfs_printcb.
Which in turn is called once for each entry read by ntfs_readdir.

Wouldn't it be nicer to allocate one 255*max_utf8_size buffer in
ntfs_readdir and use that for all entries. Assuming 255 is the upper limit
on NTFS filename length, as I read somewhere.

smbfs does something like this, except it allocates a buffer at mount time
(which works since it is only used while protected by a per-mount lock).

/Urban