Skip to contentSkip to conversation

2.5.86+: sizes of almost all files in sysfs are 4k?

DISCUSSION Alex Riesen 2 participants 3 messages Apr 23, 2003 – May 13, 2003

lore
Share & tools
lore
#0 lore raw

This confuses some applications (i.e., the midnight commander).

Was this intended?
If the size is not simple/possible to calculate, maybe using 0 would be an option for the cases where the size doesn't carry any information (like in procfs)?

-alex

#1 lore raw
Quoted 3 lines · Alex Riesen

This confuses some applications (i.e., the midnight commander).

Was this intended?

Yes. This was done at the request of one annoying konqueror user that wanted it to be smart enough to define a default handler for the text files, so he could click on it and get a preview of the contents in the sidebar.

[ I know it's a pretty poor reason to add a feature, but I caved anyway. ]

Quoted 3 lines

If the size is not simple/possible to calculate, maybe using 0 would be an option for the cases where the size doesn't carry any information (like in procfs)?

It was 0 before, which works fine for cat(1). By hardcoding the size, some bugs are exposed, since the size is reset for some reason when you try to open the file for writing, even if open(2) returns an error.

Ideally, we should be calculating the size, and using that. However, that would involve keeping type information about the file around, which we don't currently do. Research/patches in this area would be greatly appreciated.

-pat

#2 lore raw

Patrick Mochel, Tue, May 13, 2003 18:58:34 +0200:

Quoted 4 lines
Quoted 3 lines

If the size is not simple/possible to calculate, maybe using 0 would be an option for the cases where the size doesn't carry any information (like in procfs)?

Quoted 8 lines

It was 0 before, which works fine for cat(1). By hardcoding the size, some bugs are exposed, since the size is reset for some reason when you try to open the file for writing, even if open(2) returns an error.

Ideally, we should be calculating the size, and using that. However, that would involve keeping type information about the file around, which we don't currently do. Research/patches in this area would be greatly appreciated.

we could also pretend the files are sparse, and return zero-filled data. Though, i fear, this will also confuse something.

-alex