diff: mmap the file to be diffed

[pstef]

Summary

mmap the file to be diffed to prepare the land for modification of
the code to use mmap.

First component to be converted fo mmap: asciifile function to detect
if the file is a binary file or not.

diff: convert the hash to mmap

diff: Simplify the file comparison with mmap

diff: convert the check function to use mmap

[bapt]

• diff: mmap the file to be diffed
• diff: convert the hash to mmap
• diff: Simplify the file comparison with mmap
• diff: convert the check function to use mmap

[bapt]

extending the list of potential reviewers, just in case ;)

[markj]

usr.bin/diff/diffreg.c

350

Are CAP_READ and CAP_SEEK still needed?

452

Is it possible to use const uint8_t * in this function and others to make it more clear that these functions are not operating on nul-terminated strings (in the D_BINARY case at least)?

893

Can memchr() be used?

[bapt]

• diff: mmap the file to be diffed
• diff: convert the hash to mmap
• diff: Simplify the file comparison with mmap
• diff: convert the check function to use mmap
• diff: use uint8_t instead of char
• diff: use memchr to find end of line

[bapt]

usr.bin/diff/diffreg.c

350

yes not everything is done yet on the mmap, there are still rewind and fread operations in the output function, I may switch it do mmap later

[markj]

usr.bin/diff/diffreg.c

747

If we have buf2 == end2 then this dereference may not be safe, I think.

759

Similarly here you need an extra check to handle the EOF case. It looks like there are several other places below. Am I missing something?

812

Same problem here.

[bapt]

• diff: add guards against dereferencements

[markj]

usr.bin/diff/diffreg.c

711

It can just be a function, IMO.

761

Per style.9 the && and || should be before the line breaks.

804

Not sure what the XXX refers to exactly, but this is not the same logic as before. If one of c and d is EOF, but not both, then chrtran(c) != chrtran(d) is true.

812

Same problem. This does not behave the same way as before if we are at EOF on one file but not both.

[bapt]

Reduce the scope of the change to only readhash and files_differs skip check()
for now, as it is more complex and does not really bring any performance
improvement

• diff: mmap the files before actually trying to look up for differencies
• diff: simplify file comparison now we have the files already mmaped

[pstef]

Could we make this (mmap) into an option?

usr.bin/diff/diffreg.c

375

files_differ() returns either 0 or 1, so the code under default: will be never reached, I think.

510

I would rename this buf to walk and remove the object named walk below. It's a copy anyway.

[bapt]

In D28402#714447, @pstef wrote:

Could we make this (mmap) into an option?

What would be the point for it to be an option? I think if it is safe and faster we should just use that code, instead of complexifying diff code with 2 differents code path

[pstef]

In D28402#714642, @bapt wrote:

What would be the point for it to be an option? I think if it is safe and faster we should just use that code, instead of complexifying diff code with 2 differents code path

mmap is problematic whenever any of the files can be truncated. But I don't insist we make this an option, I just wanted to have it pointed out during review.

[bapt]

Addresse @pstef issues

• diff: mmap the files before actually trying to look up for differencies
• diff: simplify file comparison now we have the files already mmaped

[pstef]

Swim or surrender!