This 'sentinel' lets newsyslog know that this file is in FreeBSD UCL version 0.1. If it does not see this at the top of the file, it treats it as a legacy .conf file in the original format.

This will allow the new format to keep the original filename, newsyslog.conf
It also allows for 'versioning' of the config schema, so we can easily keep backwards compatibility if we change the config syntax again in the future.

This is a comment, of the original line from newsyslog.conf

A new format for the time specification would likely be very welcome

Working with Vsevolod Stakhov (cebka@) to come up with a way to do opaque strip splitting (https://github.com/vstakhov/libucl/issues/36)

so this could be: flags = binary, nosignal
or something similar

Dirty hack until contrib/libucl is updated. Not in a rush for this as there are some more features to be added to libucl that would be useful for this project.

taste the first line of the file. If it isn't what we want, rewind() and continue as before

Otherwise, pass it off to the new config parser

This is the map that uses the callbacks. It ended up different than I originally pictures, but I think it is still nicer than a giant strcmp() if ladder

This is mostly just verbatim from the original code. It can likely be improved, but I wanted to try to avoid introducing too many new bugs at once

Note to other reviewers: UCL doesn't implicitly understand 'mode' is octal, we just parse this field specially later in the patch.

Using double for this is odd. Usually we just multiply by a power of 10 (or 0x10).

Check the return value of fgets(3) before using line.

style(9): The body of the if() statement goes on a line by itself.

Floating point equality is somewhat fragile. Use integers for this.

You probably want to lseek(2) the fd back to the beginning — fgets(3) and other stdio functions may buffer large chunks of the file and the fd may be well past the beginning of the UCL stuff.

extra space (style)

extra space

There's lots of this, so I'm not going to mark it every time. But, in style(9) it is suggested that if/while/for braces be omitted when they are not required (with exceptions for complicated bodies). printf is not complicated.

style(9): return (true);

Ah, I see, you force octal here. Ok.

these should be on the same line (style(9))

So make 'uclver' an int, and multiply the result of strtod(3) by 10?

That'd work. You could also just use an integer version (e.g. 1) and strtol or whatever. (This is just versioning the config schema, right?)

These on the same line too

extra space

Don't need the backslash. typedef isn't part of the preprocessor.

It looks like the 3rd parameter here should be struct conf_entry * based on later functions.

You can drop the lineptr variable entirely. You would use strtod like:

uclver = strtod(&line[5], NULL);

I think you can drop the (conf_helper_func) casts if you change the function typedef from void * to struct conf_entry *.

The usual pattern for this is:

if ((working->flags & CE_NOSIGNAL) != 0) {

You can drop result entirely.

target->uid = ucl_object_toint(obj);

Use atol or strtol. uid_t is unsigned and yes, somebody has >2 billion uids (or at least uses the high numbers).

target->uid = atol(val); or similar

target->uid = pwd->pw_uid;

extra parens, you don't need them around function calls

Probably atol here too.

Probably: if (sscanf(...) != 1) {.

sscanf may return EOF (-1) as well as (0) depending on what went wrong. Anything other than 1 conversion is failure for us.

(same re: sscanf(...) != 1.)

No 'n' needed for xz, I think?

same line

also same line

same

braces can be dropped here.

same

You can drop the comment, the pattern is well understood.

Alternatively, you can include sys/cdefs.h (if it isn't already) and use __unused like:

conf_set_unknown(..., struct conf_entry *target __unused)

extra space

If you want to reduce indentation, you can do:

if (strcasecmp(...) != 0)
  continue;

And then the rest of the loop body can be unindented.

Can drop NULL check entirely. Just don't put a NULL callback in your table :-).

if (!found && ...)

(ret)

I originally had this code in libucl, so it was generalized to work in places other than just newsyslog. I just dumped it into newsyslog for now.

dropped

see response to https://reviews.freebsd.org/D1548#inline-9104

I moved the rewind(cf) that was done only in the case of a non-ucl config earlier, so it should resolve this

Not my code, but since it is inside my function, fixing it

ok

this case doesn't need to be here, it is left over. If the user entered a number, it will by type UCL_INT, unless they put a decimal in it or something silly, in which case they deserve an error anyway.

those were existing, but removed

It is done so it allows the user to do 'xzip' as well. same reason bzip is cut to 4 chars, so bzip or bzip2 both are accepted.

This is there on purpose. We have a NULL callback for the 'file' key, because it is handled elsewhere, and we don't want that key to trip the 'default' callback. See line #1486

This should be documented better, my goal is to make something generic to speed up the process of converting other utilities to UCL.

Ah, I got bzip/bzip2 but hadn't seen xzip before. Seems reasonable to me.

Ok.

Typo — I think you want !found?

This question made me think about it more, and improve it further
use strncasecmp(val, ..., 2) in all cases:

so gzip accepts: gz gzip
bzip2: bz bz2 bzip bzip2
xz: xz xzip yes
none: no none

(added yes to xz since we accept no)

good catch, thank you

Maybe just strcasecmp() for "yes"?

style(9) prefers not having init inline

theoretical overflow not handled, but is likely fine

TODO comment :)

if (!success) {} ?

if ul == 0, check errno?