« first day (1675 days earlier)      last day (2137 days later) » 

gasp
So many issues on metasmoke
 
12:51 AM
Lots of fresh new stuff to fix
 
Valid feedback types: tp fp ignore rude naa tpu fpu?
 
@SmokeDetector Custom mod-flagged: While some affiliation with pageoptimizer.pro is implicit in the josh5 promo code that's mentioned, what the actual affiliation is needs to be explicit in the post. Given the user's history of self-promotion, I suggest that a mod have a talk/inform them of SE's requirements for self promotion. Basically, this user appears to be here primarily to self-promote, not provide public answers.
 
tpu- by Makyen
 
Will anything break if MS starts only accepting tp(u) fp(u) ignore naa and rude (and the -versions)?
 
tp(u)(-)
 
12:55 AM
Yeah, I got the dash version
 
I don't think so
 
!!/watch pageoptimizer\W*(?:dot\W*)pro
 
[tag:dev-on-prod]
 
@Makyen Added pageoptimizer\W*(?:dot\W*)pro to watchlist
 
Lemme pull a list of feedback types
 
12:55 AM
arg. forgot the ?
 
Awww you're no fun
 
!!/watch pageoptimizer\W*(?:dot\W*)?pro
 
@Makyen Added pageoptimizer\W*(?:dot\W*)?pro to watchlist
 
Ambiguous	262
f	16
false	44
false-	61
false-,	2
falseu	6
falseu-	25
fp	11540
fp-	33310
fp-\	1
fp-naa	2
fp+	2
fp=	1
fpi-	2
fpo	1
fpu	175
fpu-	276
I'm being abused	1
ignore	430
ignore-	603
ignore-_shouty_case	2
ignore,	1
ignore=	2
k	3
Mith_now_youve_made_everyone_do_it	1
n	3
naa	2467
naa-	10927
naa,	2
naa?	1
naa2	2
This feature is brought to you	1
This is Smokey. I've come to say goodbye, as I've been permanently broken	1
tp	18004
tp-	5371
tpi-	9
tpu	6157
tpu-	216441
tpu--	2
tpu-.	2
tpu-]	2
tpu-\	4
 
Also, what do we want to do with the past invalid feedbacks? Nuke em?
 
12:57 AM
!!/watch bit\W*(?:dot\W*)?ly/seoproof
 
@Makyen Added bit\W*(?:dot\W*)ly/seoproof to watchlist
 
@thesecretmaster I'd leave them, just in case retroactively removing them breaks something
 
Ok, I was gonna make a "fix everything" migration
 
Dang it, what happened to the ? again
 
false is a real feedback?
 
12:57 AM
:45222044 Added bit\W*(?:dot\W*)ly/seoproof to watchlist
 
@thesecretmaster false and true are/were real, yeah
 
Objections to me normalizing them?
 
Proooobably not
 
@SmokeDetector That's not good. What happened to the ? this time?
 
true => tp, trueu => tpu, etc. (and murder the random chars that got appended for no reason)
 
12:58 AM
CI on 16fb3bf failed.
 
@SmokeDetector oddly pleasing commit hash, somehow
 
Yeah, I know. So much for being able to clean up via !!/unwatch. PR in a moment.
 
@thesecretmaster Seems okayish to me. It'd sure make that query prettier.
 
@Undo I can move the feedbacks to a legacy_feedbacks column in case we need it later?
 
That's a good idea
Easy to drop in the future, no cost now
 
1:00 AM
K, gimme 5 mins
 
Actually, why not make a not-normalized-feedback-type column, and store future non-uniform feedback there too?
That way it's easy to retroactively fix any normalization issues.
 
Because I wanted to do it with an AR validation
 
oh, and they happen at a weird spot in the pipeline
 
Yeah
Also, better if stuff stays compliant
 
Works for me
 
1:04 AM
Actually... I'm gonna add a new feedback type, 'invalid' for the ones I can't normalize. MS won't accept new feedback of that type, but it'll exist.
 
I'd appreciate someone merging PR2262, which cleans up the duplicated watch.
 
Done
 
Thanks.
 
CI on 7419170 succeeded. Message contains 'autopull', pulling...
Restart: API quota is 19623.
 
BTW: Anyone know why SD took the original version of this message a second time, not the edited version? At least in the past (e.g. when changing from !!/watch to !!/watch-force (or similar)) SD has always taken the new version, not the original.
2
 
1:17 AM
k => tp, n => naa, f=> fp?
 
@thesecretmaster There are at least these aliases.
 
I know, that's what I'm working from, but is k=tp, n=naa, and f=fp?
 
k=tpu-, n=naa-, and f=fp- There's a table there that lists them. The code probably has a better list.
 
@Makyen that issue exists with the comments.
 
@WELZ Is is just random? or are you saying that it also affects edited replies which turn into comments on MS, giving an exact duplicate comment instead of an updated one?
 
1:24 AM
@Makyen the latter
I haven't experienced (or seen it) with watch though.
Can I test it right now with a fake watch item?
For science
 
TIL: Don't edit a SD message which you really don't want to be done exactly the same twice. Easy work-around: just send a new message.
 
@Undo Running my migration+test on dev now, looks like the migration will take > a minute to run because AFAIK there's no way to bulk update stuff.
 
k, should be okay
 
Recovered from ssl.SSLError: [SSL: BAD_LENGTH] bad length (_ssl.c:2130)
 
Also, latest MS dump gave me some ASCII error
 
1:28 AM
@WELZ Unless you can fix the resulting problem yourself, or someone with the appropriate GitHub privileges agrees to be around to fix it, I wouldn't intentionally do something that you're expecting to break our normal process.
 
If you're at a dev machine, can you test it?
 
Got a branch?
 
I mean test importing the dump
I'm working off a ~1 month old dump because the new one won't import right
 
k, hold on
How are you importing?
 
I'm dumb
I'm doing it the dumb way
Forget I said anything
 
1:31 AM
It has problems with UI imports; command line works better. Bet it's that? ;)
 
[ SmokeDetector | MS ] Bad keyword in answer: Is it possible now to record screen on iOS 11? by John Ruiz on stackoverflow.com
 
Sequel Pro hates it, at least
 
Dumber
 
... trying to import the .gz?
 
1:32 AM
hah
 
Anyways, migration took 300ms on dev
 
Safari automagically unzips it, so I never worry about it
yeah, that'll be fine. Will be way slower on prod, but still probably okay
 
@thesecretmaster Smokey will never send false/true even if someone used them
 
Does MS run from db:schema:load or db:migrate?
 
Always tp/fp
That wasn't always true, however
 
1:37 AM
@thesecretmaster It migrates.
Part of the deploy process
 
k good, for some reason locally AR always converts the schema to CHARSET=latin1
 
yeah, it does that
 
@Makyen that might be a caching issue
Oh
Let me check how this event works again
shouldnt it be message.content?
 
[ SmokeDetector | MS ] Potentially bad keyword in body: Why are Jews Fascist? by user21360 on politics.SE
 
Yeah
OK I think it is caching but I see a way to fix it
 
1:42 AM
[ metasmoke ] ci/circleci success on a343566: Your tests passed on CircleCI!
 
\o/
 
Aaand I'm back!
Contributing to MS again :)
 
oooh a case
makes GH formatting unhappy
Deploying in a sec
 
takes bow
I was gonna do a massive if statment, but ew.
Hrph
Still mysql errors, but at least I know how to fix this one
 
1:45 AM
@quartata Great
 
I'll hold off until you give the okay
 
You can deploy, it ran fine. I wouldn't have pushed if it wasn't. I've got errors with the new (unzipped) dump. But those I can fix.
 
[ metasmoke-deploy ] deploy started by Undo
 
Incoming
 
1:47 AM
sd f
 
[ metasmoke ] ci/circleci success on 20b56e2: Your tests passed on CircleCI!
[ SmokeDetector | MS ] Bad keyword in title: "Who lives there?" vs "Who live there?" by Linguisticpasca on english.SE
 
@SmokeDetector @Undo Is this really necessary?
 
The message?
 
No, the commit
Because won't prod just overwrite my changes to the schema?
 
Somehow it doesn't do it on prod
Might be something about the MySQL configuration being different.
 
1:52 AM
Yeah, I think on prod it'd just overwrite schema.rb when it migrates with unicode
 
I've never thought about whether it's necessary, though. It might not be.
 
I think it is. Art and I had an issue with emojis (?) and latin encoding in mysql while working on SCOT.
 
The encoding is necessary, but the changes to the schema file might be irrelevant.
 
db:migrate == db:migrate; db:schema:dump IIRC
 
fp- by Catija
 
1:54 AM
Huh, TIL
 
I could be wrong, lemme check
 
@SmokeDetector So... why is "who live" a filter?
 
Recovered from ssl.SSLError: [SSL: BAD_LENGTH] bad length (_ssl.c:2130)
 
@Catija we need better history on that kind of thing
It's a huge project to take on, though
 
So you can't see who added the regex for it?
 
1:56 AM
We can, if we can find the regex
There's an active issue for adding a non-manual way to do that
 
Ah. So Smokey can say what triggered it but not which chunk of regex was triggered.
 
Right
Regexes are awesome, but they're not really good at telling you why something matched. Just that it did.
So you're basically stuck with seeing what content matched (which we have), but it's hard to see what part of the regex it matched.
Have to do some binary search magic, AFAICT
 
> it will also invoke the db:schema:dump task, which will update your db/schema.rb
Sure enough. Interesting
That'd explain it
 
The schema is only actually read if your run db:schema:load (or that's called through something else, like db:setup)
But it's only really used when e.g. you clone a repo and want to setup a db from scratch
 
2:13 AM
@SmokeDetector ignore
 
@Makyen Post ignored; alerts about it will no longer be posted.
[ SmokeDetector | MS ] Offensive answer detected, toxic answer detected: Can "face pulling" or "bone pulling" really move/remodel bones in adults? by user43364 on biology.SE
 
Wait what
I did not realize that there were actual feedbacks that were "Ambiguous"
 
2:31 AM
7 hours ago, by Zoe
It's probably to catch some of the posts promoting "watch someteam vs someotherteam live"
 
[ metasmoke ] ci/circleci success on 49e198b: Your tests passed on CircleCI!
 
@SmokeDetector k
I'm planning to make a command that shows versions of all dependencies
 
Something like !!/pipver and it says aaa==1.2.3, bbb==2.3.4
 
@quartata kaboom
 
2:35 AM
I plan to parse the output from pip freeze and filter to only show Smokey's requirements, so other packages won't be revealed.
 
meh, try git blame.
 
A certain Jed Fox?
 
Huh... does FIRE update to reflect the deleted/non-deleted state of a post?
 
it was Undo
 
2:37 AM
I feel like I've heard that name
 
@thesecretmaster he only touched that line to remove the return
 
Oh
Well, I'm gonna kill 2/3 of them
 
wait, return before end?
 
@Catija it should show with the deleted background after its gone, yes
 
@iBug Yes?
 
2:38 AM
There's one in the tavern doesn't...
 
@thesecretmaster I think the return keyword should've been dropped...
 
metasmoke probably doesn't know it was deleted.
 
@iBug Yeah, JedFox dropped that
 
Shog deleted it... not sure how... probably just normal deletion.
 
Oh, I was looking at an old revision.
 
2:39 AM
 
[ SmokeDetector | MS ] Luncheon meat detected, no whitespace in answer: How to use Rowbounds class in Mybatis mapper.xml? by K K on stackoverflow.com
 
@SmokeDetector k
@SmokeDetector The username says it all: K K!
 
@SmokeDetector why
 
@Catija Body - Position 769-777: kill you
 
2:55 AM
@Catija Ummmm... You're using FIRE. Are you aware that within the FIRE popup the "why" information is displayed in a tooltip-like display when you hover over the post's title?
 
@Makyen No. How would I know that? It's pretty non-obvious.
 
[ SmokeDetector | MS ] Few unique characters in body, potentially bad keyword in title: Memes hahahaahahah by Ya mom on blender.SE
 
@SmokeDetector k
 
@SmokeDetector I also do
 
@Catija OK. Now you know. :-) Having just checked, it doesn't appear to be in the documentation, and should be. I believe I discovered it when moving the mouse to the action buttons, but I do agree that It should be more clear/documented.
 
3:02 AM
Thanks :D Sorry, that came out a little more harshly than I meant to have it be.
 
np. I, also, am sorry if anything I said came out as anything other than just trying to be helpful.
 
:D
 
Anybody have a userscript to switch all fonts on a page that are "A font" to "Arial"?
 
tp by Nisse Engström on Do I need any drivers? [MS]
tp by Nisse Engström on trghtehthethhhhhhhhhhhhhhhhhhhhh [MS]
 
!!/watch (?:corn)?i love corni?
 
3:10 AM
@Makyen Added (?:corn)?i love corni? to watchlist
 
!!/unwatch i love corn
 
@Makyen Removed i love corn from watchlist
CI on db7117d succeeded. Message contains 'autopull', pulling...
[ SmokeDetector ] SmokeDetector started at rev db7117d (SmokeDetector: Auto unwatch of i love corn by Makyen --autopull) (running on teward/Solar Flare)
Restart: API quota is 19049.
 
@SmokeDetector ignore
@quartata The issue of my feedback, ignore, being eaten as a comment just happened again.
 
3:44 AM
[ SmokeDetector | MS ] Bad keyword with email in answer: Petrov Defense, easy equality for Black? by Longview on chess.SE
fp- by tripleee
 
really weird though; flagged for moderator attention
@WELZ this doesn't sound like a natural fit for Halflife but if you want to open a ticket, it's probably the best place we have for that right now
the current design is that Halflife has a git clone of SmokeDetector so it can stay up to date with blacklist additions
it's not particularly elegant (and it could be improved; currently it still just does a pull in the blind I think every 10 minutes; it should use the Github API to get a notification when the code receives an update) but it's probably a good place to start
 
[ SmokeDetector | MS ] URL in title, bad keyword in body, bad keyword in title, blacklisted website in body, blacklisted website in title, +4 more: freemp3download.xyz/dumb-shit-rayy-dubb-mp3-song-download/ by elxyovf on english.SE
tpu- by NobodyNada
 
another option might be to revive Quartata's code to identify why exactly something was flagged by a regex, but make it optional, i.e. !!/why would normally not tell you this but you could say !!/wtf to activate that code and rerun the analysis, and receive back a more detailed report
 
@tripleee MS sends notifications on git update tho
It's available to the API subscribers now not just SmokeDetectors
You made that FR I think
 
yeah, there are several things that Halflife wanted which I should just pick up
do you think putting this !!/wtf code in Smokey could fly, though?
do I remember correctly that you had a version which did this already?
 
4:00 AM
It would be fairly easy, but having two regexes in memory kinda stinks
The BlacklistFactory is pretty easy to use and fast it just wasn't fast enough (although it probably would have been OK, just not when combined with Perspective)
I might bring it back for every scan anyways after this refactor if I can boost performance in other places
 
so this could be in Smokey after all?
 
@quartata do we need to keep compiled regexes in memory? even if it takes 5 seconds to run through all of the blacklists, that's not a huge deal if it's manually invoked
 
That's fair
The BlacklistFactory caches a mapping between capture groups (which each one is wrapped in) to the blacklist item info
 
the really tricky ones to track are often the hardcoded ones, though
 
This included the nwb ones
 
4:05 AM
@tripleee if I implemented a !!/wtf command, I'd probably run through those to -- though that couldn't go in the BlacklistFactory
@quartata There's even more special cases though
 
like the e.m.b.e.d.d.e.d...p.u.n.c.t.u.a.t.i.o.n one we tracked down last week
 
But not the other rules, admittedly
@tripleee well ok that's a fault of people lacking good reason names
Ultimately when we do the big configuration refactor it'll probably be easier to track these things down, since items will be categorized
(we could use the BlacklistFactory there for the watchlist)
 
@quartata Some of them are e.g. blacklisted keywords that should be restricted to only a few sites, or only titles, etc etc
 
a third option is to call this "offline hygiene" ... I suspect we have around 20% regexes which used to be useful but which no longer catch anything or only produce FPs
 
Especially for watches, since it kinda needs to be "Potentially bad keyword in {}" to be a watch
 
4:07 AM
but in these monolithic lists of hundreds of regexes you can't pick those out individually
so you could run the code base over a copy of the Metasmoke archive periodically and see that "this one stopped being useful in September 2016"
or "this variation of 'casinos in Delhi' or 'escorts in Bangalore' actually never caught anything"
 
That is what @angussidney's scripts do
 
we have that? where?
 
A little tweaking is needed but
 
like a README (-:
mmm but again that's just the blacklist, not the entire code base
it's the hardcoded mess in findspam.py I want to see cleaned up
in other news, where are the pharma spammers today? they seem to be losing interest in us
15 hours ago, by SmokeDetector
[ SmokeDetector | MS ] Bad NS for domain in body, bad keyword in body, blacklisted website in body, link at end of body, pattern-matching product name in body: fat cells in the body until the finish of time by Brazilboyzs on superuser.com
this is still the latest
 
[ SmokeDetector | MS ] URL in title, bad keyword in body, bad keyword in title, blacklisted website in body, blacklisted website in title, +2 more: youtubetomp3converter.world/waiting-yellow-claw-mp3-song-download/ by dorcogp on english.SE
tpu- by NobodyNada
 
4:23 AM
BTW I got a response from DataTurks regarding the spam yesterday, it was a junior intern metasmoke.erwaysoftware.com/abuse/reports/27
 
New metasmoke user 'Doorknob' created
 
welcome, mighty @Doorknob
 
@tripleee \o/
 
@tripleee not sure where you got "mighty," but thanks :P
 
100k+ rep is always impressive
but really it just kind of rolled off my tongue
 
4:30 AM
gotta be honest, I've never gotten that one before
 
it's a memorable user name alright
another long-time favorite of mine is Hostile Fork but I haven't seen him around much
 
@tripleee to be fair, I was really young when I came up with it
@tripleee Dr. Rebmu! they use to be fairly active on code golf
 
judging from your profile, you still are ... but probably older than several of the regulars here
 
[ SmokeDetector | MS ] Bad NS for domain in body, bad NS for domain in title, bad keyword in body, bad keyword in title, blacklisted website in body, +4 more: Konect Nutra Keto www.healthcareorder.com/konect-nutra-keto/ by Agneansfr on askubuntu.com
tpu- by tripleee
 
@tripleee (that's a strange feeling... it's been too long)
 
4:36 AM
welcome back #drugs spammers! we missed you
 
Restart: API quota is 18552.
 
they are very late, they started at 3:53 yesterday
 
sd why
 
[:45223284]
Blacklisted user - blacklisted for //magento.stackexchange.com/a/230169 (https://m.erwaysoftware.com/posts/uid/magento/230169) by the metasmoke API
fp- by A J
 
4:43 AM
Hmm, might actually have been spammy originally
 
the one they were blacklisted for was way more aggressive, they are probably slowly learning
 
It somewhat resembles his original spam: "traditional SEO advice is worthless, I know what you should do instead but I won't tell you on-site"
 
how on earth did @Doorknob not already have an MS account
 
@quartata heh, smokey sure has come a long way from the hacked-together websocket watcher it used to be
 
@Doorknob ...the original userscript was yours, wasn't it?
 
4:48 AM
it was :D
I'm glad to see the "sd X" feature is alive and well, that's one remnant of my contributions :P
 
if you need more Metasmoke privileges, I think it's just a matter of asking ... the SQL URL I posted above requires Core privileges IIRC
how can I configure the name of the feed user? It's just Feeds in Charcoal Test and I don't seem to find a way to change it
 
[ SmokeDetector | MS ] URL in title, bad NS for domain in body, bad NS for domain in title, bad keyword in body, bad keyword in title, +4 more: www.usahealthmarket.info/vivrax/ by user122736 on graphicdesign.SE
tpu- by NobodyNada
 
@tripleee thanks, I'm just stopping by out of curiosity though
 
@NobodyNada They've had 3 posts reported. 1st & 3rd (both deleted)) were definitely over the line. The 2nd, this one, wasn't that bad, but was still suggesting contact outside SE. The 1st's been described. The 3rd had 2 obfuscated URLs, one of which with a promo code with the user's name in it. I've custom mod flagged, twice (both helpful). This most recent edit, by a mod, was probably one of the results.
 
@tripleee I think you have to create a new feed user and assign it to the feed, and then you can edit the user from there
 
4:57 AM
@Doorknob do you have to be a mod to do that?
 
oh, maybe, one sec
 
nvm, just wondering how it was done here when Processed Meat was created ... I wanted to call mine Planet Pharma but it can just be Feeds for this brief experiment anyway
my subconscious woke me up with this idea this morning, look for RSS feeds on the spammer's blogs and add them to a feed
maybe down the line if it actually triggers when they add new posts, see if we can predict what they are going to start spamming next
 
do you see "create user" buttons when you click the "feeds posted into this room" link?
@tripleee haha brilliant
 
I only see "manage feeds" in the room dropdown, where should I look for that?
 
[ SmokeDetector | MS ] No whitespace in body, blacklisted user: AGFDHUIRGHUAFADDSFASDF by Ya mom on blender.SE
 

« first day (1675 days earlier)      last day (2137 days later) »