mozilla :: #activity-stream

8 Aug 2017
01:00dmoseMardak: it is not a WIP; i should have gotten rid of that. i usually manually wrap comment lines, yes.
06:43streaminatordmose: Hey, your patch was approved!
06:43streaminatordmose: Hey, your patch was approved!
15:18ursulaMardak: i thought the use for AS in the query was so you can reference it like a variable name later in the query
15:19Mardakyeah you can use it for referencing from the query too
15:20Mardaki would think the getting a column's value by name is using the same name lookup
15:24ursulaMardak: sorry, i'm a little confused then. why do you need moz_bookmarks.guid AS bookmarkGuid but not moz_places.guid AS guid, then?
15:24Mardakbecause without AS, there would be two "guid" columns
15:25ursulacuz moz_bookmarks and moz_places both have guid
15:25ursulai see
15:25ursulaok feel free to close those issues then
21:08streaminatorursula: Hey, your patch was approved!
21:15dmoseMardak: so at this point, i'm inclined to do the postMessage thing
21:16Mardaki get these results:
21:36Mardakdmose: there's some scary bolded stuff on
21:37Mardakmaybe a good enough check is to make sure the event.source ===
21:37Mardakor that event.source === window
21:37Mardak(they should all be the same in our usage i believe)
21:48Mardakursula: TIL horizontal dots = meatballs menu vs vertical dots = kebab menu :p
21:48Mardak9 dots = bento menu ?
21:49ursulaPhoton lingo!
21:53Mardakand for those confused by the dner menu:,_Istanbul,_Turkey.JPG
21:56dmoseMardak: yeah, boy. i'm inclined at this point to just go with rAF + setTimeout
21:57dmoseon the theory that digitarald posited: it should be good enough. plus it's let complicated
21:57dmoseMardak: it's not particularly clear to me that how much we care what the paint times of backgrounded stuff is.
21:59dmoseMardak: i mean, harald's suggestion of not even collecting those may be worth diong, though I don't think we want to do it as part of this patch.
22:04Mardakdmose: load_trigger_ts will be set on a preloaded browser of the current tab open command
22:05dmoseMardak: sure. and?
22:05Mardakre: 500k milliseconds to load
22:05dmoseMardak: well, no. load_trigger_ts doesn't get set until somebody does something to trigger it to be seen.
22:06dmoseMardak: eg cmd-T
22:06Mardakyes.. open a new tab with cmd-t. mark is set
22:06Mardakthe preloaded browser will use that mark
22:06dmoseright, and...
22:06dmoseshould we vidyo this?
22:06Mardakthat's why emtwo's graph has 500k milliseconds to load
22:07Mardaker. 500k milliseconds from load to visibility
22:08dmosewait, you're saying that the preloaded browser will using the load_trigger_ts from the previous new tab?
22:08Mardakwell "previous" is unclear but basically yes
22:08dmoseas far as i know, that can't happen
22:08dmoseoh, but maybe it could
22:08dmosei bet you're right
22:08Mardakif there was no preloaded browser and user hits cmd-t the first time
22:09Mardakthe unpreloaded page gets the mark
22:09Mardakand the preloaded browser also gets the mark
22:09dmoseok, so that's separate from the stuff we're otherwise talking about, i think
22:09Mardakif there is a preloaded browser, it will have the mark from the previous tab opening
22:09dmosethough also worth fixing ASAP
22:10dmoseMardak: so, first, what do you think about my reasoning above re setTimeout?
22:10Mardakit comes back to what are we actually trying to measure?
22:10dmoseyes it does
22:11Mardakif we don't care about hidden/preloaded stuff then the solution is a bit different
22:11dmosewe don't care about hidden/preloaded _paints_
22:11dmosei think we care about other stuff like visibility
22:11Mardakwell solution of setTimeout would probably be acceptable
22:11dmoseas far as what we do about that double usage of the mark, thing, i need to ponder that a bit
22:12dmosethe medium term fix is that we should probably have an identifier associated with each browser
22:12dmoseand append that to the mark
22:12Mardakit is giving us some metric of roughly 8 minutes between newtabs in the 75th percentile. not sure if that is useful but it is something measured ;p
22:12dmoseif we want that, we should measure that explicitly :-P
22:14dmoseso, at some level, this is tied up in figuring out if a browser is preloading or not
22:14* dmose pokes around
22:14Mardakwhich we can tell via visibility change, yeah?
22:15Mardak if (this.document.visibilityState === VISIBLE) {
22:15dmoseat certain points, yes
22:15dmoseand that's probably enough for what we need now
22:16Mardakand if we get a visibilitychange, to visible, it means a preloaded browser is now showing. so *now* set load_trigger_ts
22:16Mardakbased on the most recent mark
22:16dmoseso maybe all we have to do is check the visible flag in setLoadTriggerInfo
22:17dmoseand if it's set, return early
22:17dmosealthough that's chrome code
22:17dmoseand so hmmm
22:17Mardaknod and visibility is in content
22:18Mardaka related issue that emtwo needed to work around was the topsites painted data happening before visibility
22:18Mardakwhere it could be fine as it is right now
22:18dmosewhich might be a good reason not to collect that at all
22:19dmosewhat do you mean?
22:19Mardakif topsites before visibility -> implies topsites are actually painted at visibility time
22:19Mardakor we could just send the topsites one on visibility
22:19dmosethat too
22:20dmosehere's a slightly hacky fix:
22:20dmoseactually, that won't work
22:20dmoseMardak: vidyo?
9 Aug 2017
No messages
Last message: 13 days and 7 hours ago