libxo'ify netstat.

[alfred]

Summary

Original diff: Phil Shafer
Additional code: Kim Shrier <[e-mail]>
Sponsored by: Norse

[alfred]

• remove WIP files

[alfred]

• Cleanup route json.
• libxo p_rtentry_kvm
• Fixup route output from kvm

[marcel]

Thanks!

[ngie]

I glossed over the review. There are a lot of changes and I'm not aware of how the libxo grammar works, but if this could be done in phases it would make it a lot easier to review...

usr.bin/netstat/Makefile

45–46

Use LIBADD instead of DPADD/LDADD? Just watch out for MFC issues..

usr.bin/netstat/if.c

129

xo_open_list and xo_close_list can fail. I think we care because that means that the output gathering would be all for naught, correct?

This comment applies to other areas where this pattern's applied as well.

134

style(9): remove space between (uintmax_t) and *a; this comment (spurious space between cast and variable) applies to other additions as well.

254

style(9): bad indentation -- 4 spaces after hard tab (here and elsewhere).

326

This indentation looks off. Is it a phabricator bug?

331

I understand what you're doing, but this seems a bit gratuitous, and ifa->ifa_name is a bit easier to grok than name.

416

Should this be size_t?

422

Indentation, and this also seemed like it should have been a general purpose function elsewhere..

usr.bin/netstat/inet.c

352

This seems a bit gratuitous (along with the next line's whitespace change). Should this be a separate commit?

usr.bin/netstat/main.c

864

(Random question) should usage messages be converted to libxo...?

[alfred]

• use LIBADD instead of LDADD/DPADD

[alfred]

• Need to close socket instances for 'netstat -a'
• Quote pointer values.
• Fixup socket output, use a single list.

[alfred]

• Pointers need to be quoted.

[alfred]

• Remove whitespace from tcp/toe46 values.

[alfred]

• Fix protocol whitespace in -Q output

[alfred]

• Fixup padding between CPU&Name for -Q output
• Only open socket list when doing control block print
• quote icmp-address-responses
• Need to close input-histogram
• hex address needs to be quoted
• Tests for json from netstat
• Remove stray diff markers
• debug macros
• Workaround {P: can't take 0 pad and number args seem broken.
• Deal with zero lenth LINK address.
• Need to close interface instance.
• Emit well formatted json when -q given with -w

[alfred]

• Fix -B (bpfstats) with json

[se]

This comment was removed by se.

[se]

usr.bin/netstat/inet.c

1233–1234

Typo: output-histagram should be output-histogram

[se]

usr.bin/netstat/route.c

918–922

The format generated in "buffer" is suitable for printf(), but not for xo_emit().
Reason: "%s" in xo_emit() is printed verbatim, unless written as "{:/%s}" ...

The correct format strung for xo_emit() might be generated by:

snprintf(buffer, sizeof(buffer), "{d:/%%-%d.%ds }", wid_flags, wid_flags);

(untested)

[alfred]

• spelling: histogram
• Proper formatting for flags.

[alfred]

In D1380#28, @se wrote:

Typo: output-histagram should be output-histogram

Fixed!

[alfred]

In D1380#29, @se wrote:

The format generated in "buffer" is suitable for printf(), but not for xo_emit().
Reason: "%s" in xo_emit() is printed verbatim, unless written as "{:/%s}" ...

The correct format strung for xo_emit() might be generated by:

snprintf(buffer, sizeof(buffer), "{d:/%%-%d.%ds }", wid_flags, wid_flags);

(untested)

I took another direction with this, can you see if it makes sense? I now output the old flags as well as the new flags. Maybe there is even a better way to make "pretty_flags" into a key-value dict?

[se]

In D1380#33, @alfredperlstein wrote:

In D1380#29, @se wrote:

The format generated in "buffer" is suitable for printf(), but not for xo_emit().
Reason: "%s" in xo_emit() is printed verbatim, unless written as "{:/%s}" ...

The correct format strung for xo_emit() might be generated by:

snprintf(buffer, sizeof(buffer), "{d:/%%-%d.%ds }", wid_flags, wid_flags);

(untested)

I took another direction with this, can you see if it makes sense? I now output the old flags as well as the new flags. Maybe there is even a better way to make "pretty_flags" into a key-value dict?

Do you really need the textual version of the flags when generating XML or JSON?

I do actually consider the previous version better (with tag "flags" instead of "pretty_flags"). The individual elements are not "pretty", they are just the XML/JSON way of making the flag easily accessible for testing in scripts.

There is no better way to generate the key-value dict that I know of.

The "l" modifier of xo_emit() is undocumented except in the libxo sources, but it seems to be there exactly for your use case. I'd like to use a different API for libxo, which I'm going to discuss elsewhere.

(To give an idea of what I'd rather use instead of the "l" modifier: xo_emit("{:tag/...}") within xo_open_list("tag") should imply the "l" modifier for JSON output - this modifier is a NOP for all other formats, anyway.)

[• rodrigc]

Looks OK. In a future pass, the testing logic should be converted from Makefile logic that is in there now
into a test that generates a Kyuafile under /usr/tests.

[marcel]

conversion complete