http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5545
Summary: switch 3.3.x updates to require "tflags publish"
Product: Spamassassin
Version: 3.2.1
Platform: Other
OS/Version: other
Status: NEW
Severity: major
Priority: P5
Component: RuleQA
AssignedTo: [email protected]
ReportedBy: [EMAIL PROTECTED]
quoting from a thread from a month ago:
me: 'We currently have this auto-publishing of rules from sandbox to sa-update,
based on how well they do in some automated testing. The thing about
URIBL_RHS_URIBL_BLACK (which is a test rule) made me think, though --
should we modify this to require a more explicit sign-off for the rules
that we want published?
We can already do this (labouriously) by adding "tflags nopublish" to
every rule, or renaming them to have the T_ prefix. What I'm thinking
though, is that rules in sandboxes be implicitly considered "nopublish"
for sa-update use, unless *explicitly* marked "publish".
This would be in addition to the automated testing step, too. In other
words, a rule would have to:
- be in rulesrc/sandbox/whoever/foo.cf
- not named T_SOMETHING
- be listed with "tflags publish"
- pass the QA freqs thresholds
to make it into sa-update.
This was we wouldn't get test rules like URIBL_RHS_URIBL_BLACK (which
seemingly had good enough freqs to be published) getting into updates;
whereas when we write new rules that *are* intended for updates (assuming
they work and catch enough spam), they'll get published easily.
I think that should cut down on the danger of test rules getting
published when we don't want that to happen.'
Kevin A. McGrail: 'As discussed prior, I *thought* the sandbox was a playground
aka wiki sandboxes where it was just lines in the sand washed away by the next
tide. I was very surprised to find out they auto-promoted. I would support the
explicit nature you refer to below.'
Sidney Markowitz: 'I agree. I also assumed that the sandbox was set up like that
until I had occasion to run some test rules. Your proposal makes the whole
process take fewer steps as well as being fail-safe -- There's no need
to remember to use special names or flags when you are first creating a
test rule, and just the step of adding one flag when you are ready to
publish.'
me: 'Yep, that's the idea.
I forgot to mention an additional detail though -- if a publishable meta
rule relies on a testing subrule, it'd have to bring in the subrule
into the publishable set. (That's what the current code does now anyway,
so there isn't much change in that regard)'
Daryl: '+1. I think we actually discussed doing this before, too. At least I
remember sending mail about it.'
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
