possible regression going from 0.8.2 to 0.9.0

[dlangille]

I noticed this regression after upgrading from 0.8.2 to 0.9.0 today. The following is a simple example broken down from one of my use cases.

with 8.0.2:

% /usr/local/bin/ucl_tool --in test.ucl --format json
[
    {
        "type": "install",
        "message": "1) add webcamd_enable=\"YES\"\n\nto your /etc/rc.conf"
    }
]

With 0.9.0, we have additional commas:

% /usr/local/bin/ucl_tool --in test.ucl --format json
,
[
    ,
{
        "type": "install",
        "message": "1) add webcamd_enable=\"YES\"\n\nto your /etc/rc.conf"
    }
]

Source:

% cat test.ucl
[
{ type: install
  message: <<EOM
1) add webcamd_enable="YES"

to your /etc/rc.conf
EOM
}
]

[Crest]

I've been bitten by the same problem and 'blamed' uclcmd in FreeBSD PR #277666 (updated to link to this issue report).

The problem makes it impossible to use ucl_object_emit() YAML, JSON and compact JSON output format for anything but scalar UCL types (int, float, string boolean, time, and maybe user-data depending on the user-data). It breaks both arrays and objects. The YAML breakage is more subtle because of its significant whitespaces.

[Crest]

There is a fix for the emitter that was committed to the master branch after the latest (0.9.0) release: https://github.com/vstakhov/libucl/commit/da7aacb3d46fbc206ec6f2431f503eed843a3039.

[vstakhov]

Haven't I tagged that yet? Let me check...

[Crest]

I've already submitted a (FreeBSD) bug report with a patch against for the libucl port, because I didn't expect such a fast turnaround. I especially worry about what the YAML output could be parsed as because I'm not sure all outputs from libucl 0.9.0 are either syntactically invalid or preserve the document structure.

[vstakhov]

I have tagged the release 0.9.1 with the fix.

[dlangille]

Confirming, 0.9.1 fixes the issue for me. Thank you.