Monday, July 6, 2015

Beta 1 aftermath

So let's sum up the first 38 beta.

Confirmed bugs: Facesuck seems to be totally whacked in Ion mode (it works fine in Baseline only mode). IonPower passes all the JIT tests, though, so this must be something that the test suite does not cover. I'm investigating some other conformance test suites and corrected a couple other variances so far between Baseline and Ion but none of them appear to be what's ailing Faceblech yet.

Also, we have another web font that makes ATSUI puke, except it has an inconveniently null PostScript name so we can't filter it with the existing method. Fortunately Tobias had come up with an alternative font filter system some time ago that should work with 10.4.

Not confirmed (yet?): a few people have reported that memory usage skyrockets upon quit and the browser crashes (inevitably after exceeding its addressing space), on a variety of systems, both 10.4 and 10.5. I can't reproduce this on any of the test machines.

I need to do more looking into the stored passwords question.

Since we're out of runway, i.e., ESR31, and we need one more beta before release, I'm going to keep working on the Facebork problem (or at least try to fix it by fixing something else) until July 24. If we can't do it by then, I guess we launch without IonPower, which is unfortunate and will regress JavaScript performance, but we will still at least have Baseline. Faceburp is just too big to debug in place, so I need you folks to find another site that has similar problems. I haven't been able to yet myself.

40 comments:

  1. Cameron,

    I don't have a PowerPC Mac but your blog posts are a very interesting read. It's interesting seeing just how far you can push older hardware into the future.

    You're very good at tech writing, please keep these posts coming!

    ReplyDelete
  2. No other site with the facebook problem found so far.

    and never ever seen the memory problem (Powerbook G4 12, 1.5Ghz, 1.25GB)

    ReplyDelete
    Replies
    1. There must be something in common with the systems that crash, but I can't find it yet. I'm now starting to wonder if a kext or some other system extension is conflicting.

      Delete
    2. I can confirm that 38 causes this memory problem.

      17" PowerBook G4 1.0Ghz, 2GB ram (A1013) and PowerMac G4 Quicksilver, 1.2 Ghz (Sonnet upgrade), 1.5GB ram.

      Memory use will rocket upon quit. It's brought my PB to a lockup condition before I figured out what was going on. Eventually it does release, but it can take a while.

      I do have a "restart" button as well in the browser. Clicking on that does not restart the browser because it crashes out (from the memory thing) before that happens.

      Delete
    3. Chris' diagram below is useful. I'm away from my Macs at the moment but I intend to test that when I get back.

      Delete
  3. 38 beta is much, much faster than 31, especially on loading. 31 took up to 2 minutes to load on complex sites; annoying. 38 takes 10 seconds; not annoying. Using 7450 on a PB G4 1.67 GHz Hi-res. But it crashed on a PB G4 1GHz Titanium. No other problems encountered. ( I don't do facebook on account of privacy issues).

    ReplyDelete
    Replies
    1. Is this the quit-crash mentioned above, or another type of crash? If the URL is reproducible (it always crashes on this page), please advise.

      38 is much quicker because of the full IonPower suite, so I would really like to ship with it if I can.

      Delete
    2. It was a quit crash. Not sure about memory but as the Titanium is maxed out at only 1 GB RAM that could be it. On the Titanium it's still possible to use version 24.7 without the annoying delays in loading complex pages, if that helps. The Hi-Res PB G4 has 2 GB RAM. Please don't ship 38 w/out the speed improvement. Thanks for all you are doing.

      Delete
  4. Regarding the exit crash, consider this chart, tested on 10.4, Pismo, old OS X install

    http://s9.postimg.org/uzqf8jka7/crash_on_exit.png

    Meanwhile, the G4 PowerBook is preparing a new 10.5 install on an external drive to see if I can reproduce it on a clean system.

    ReplyDelete
    Replies
    1. This is *very* helpful. Thanks. What this tells me is actually this may be related indirectly to the Facebroke bug.

      I just discovered what may be the root cause today -- a major issue when JavaScript functions are inlined by Ion (Baseline doesn't use this code). Strangely, this must only occur under certain specific situations because it is not uncovered by the JIT suite. I don't know how to fix this yet, but this is potentially fundamental enough that it may be the problem. I haven't decided if this is Mozilla's bug or ours yet, but we'll have to fix it either way.

      Delete
    2. Maybe this wasn't so helpful after all… First: I can reproduce the crash on a completely new 10.5 install as well (using my usual Firefox profile). However, I was unable today to reproduce it with a fresh Firefox profile. After seeing the crash with a fresh profile initially (which may have been a coincidence), I did the rest of yesterday's testing with my usual profile, which was probably a mistake.

      So I investigated further today. I can still reproduce the crash with my usual profile with all add-ons disabled, and in safe mode. So it must be something in the profile that's still active in safe mode, but doesn't exist in a fresh profile.

      I deleted all user prefs and wasn't able to reproduce the crash in many tries. Since the crash doesn't occur every time, I found it impossible to determine which preference is the culprit even with the "first half vs. second half" iteration method in a manageable time span. Sigh.

      Delete
  5. Interesting reading. Very nice to see that some devs are doing stuff just as wizardry, hehe.

    Must be a pain in the ass to debug in a PPC.

    ReplyDelete
  6. I have been using 38.0.1 on Linux (via Iceweasel) and the TenFourFox offering on Leopard. Only 2 add-ons - noscript and bluhell firewall.

    My experience so far is 100% positive for what I do; with both.

    38 is a winner in my book.

    ReplyDelete
    Replies
    1. One thing I forgot to comment on...

      I don't do much web on Mac OS anymore, but one big improvement I have seen in TFF is the current lack of idle CPU use. It's so much better now it's ridiculous.

      Back in the ver. 17 days it would be up around 15% or more when doing nothing with no type of script or anything else running. Now it's 4% or even off the scale.

      Efficiency is everything on these older PowerPC systems, and you're clearly evolved at that. Respect.

      Delete
  7. I'm really enjoying using 38.0.1b1, but have run into an issue with the FireFTP add-on. I am unable to connect to box.com's FTP server using FTPS protocol. Initially I thought the problem was unique to TFF 38.0.1 (with fireftp-2.0.24), but I discovered that TFF 31.8.0 (with the older fireftp-2.0.19) is also problematic. TFF 31.7.0 (with fireftp-2.0.19) and Firefox 38.0.5/39 (with fireftp-2.0.24) work fine and log the following messages when I connect to ftp.box.com:

    220 Service ready for new user.
    AUTH TLS
    234 Command AUTH okay; starting TLS connection.
    PBSZ 0
    200 Command PBSZ okay.
    USER
    331 User name okay, need password for .
    PASS (password not shown)
    230 User logged in, proceed.
    FEAT
    etc.

    When I attempt to connect using TFF 31.8.0 or 38.0.1b1, only the following messages are logged and the connection attempt fails quickly:

    220 Service ready for new user.
    AUTH TLS
    234 Command AUTH okay; starting TLS connection.
    PBSZ 0

    Maybe this is related to the password manager issue that has already been discussed? Whether you create a stored account or use FireFTP's QuickConnect feature, it appears that it attempts to store the password for the account in the security device.

    ReplyDelete
    Replies
    1. I don't have any explanation if it's occurring on both 31.8 and 38.0. I try to support add-ons as best as possible, but it's best-effort, not guaranteed.

      Delete
    2. Understood. Certainly you're doing plenty without trying to support every third-party add-on! I was surprised that the behavior was similar in TFF 31.8 and 38.0.1. I started to think maybe a change in TLS handling was involved, but the fact that Firefox 38/39 works seems to rule out that possibility. Thanks for all of your efforts, Cameron. You rule!

      Delete
    3. In case anyone runs into this problem, I documented the fix here:

      https://github.com/mimecuvalo/fireftp/issues/143

      Delete
  8. One question, why does Mozilla 39/intel/OSX 10.7 uses maximum of 600/700 MB Virtual memory (4 tabls open, including facebook and the verge..heavy sites), and TenFourFox goes quickly to 1.2GB and then up to 1.4GB?

    Maybe this does not affect speed, but given real memory of 1.25GB (my case, with PBook G4) maybe it does.

    ReplyDelete
    Replies
    1. I don't know, because I can't reproduce it. On this Quad G5 with about 40 tabs open, I'm using 485MB of RAM.

      Delete
    2. this is what I get.

      Any info that can be useful to you ?

      https://flic.kr/p/v19FgS

      Delete
    3. Not really, because I believe you when you say it's taking up that much space, and that's really all your screenshot says. However, without any realistic means to replicate it on a test system, it's not something I'm prioritizing until I get reliable steps. Meanwhile, if someone wants to submit a patch, I'll certainly take it.

      Delete
  9. I'm noticing a problem in Preferences --> Security --> Saved Passwords. In 31.8, it displays my passwords, but in 38.0.1 it displays none, even though it's the same profile.

    On the memory front, everything's stable. RAM usage is about the same as 31.

    Thanks again for supporting the "Tiger Forever" crowd :)

    ReplyDelete
    Replies
    1. I can confirm the saved passwords problem. The passwords do work, so it's a display issue, but I can reproduce it here. I'd like to fix this before release, though this doesn't quite reach show-stopper status.

      Delete
  10. It's the "font that makes ATSUI puke" problem. I have a fix in hand for beta 2. (It also affects 31.8, but I won't be respinning that.)

    ReplyDelete
  11. What's funny is on my Mac Pro OS X 10.6, that link causes FF 39 to freeze for 30 seconds and generate a script error.

    ReplyDelete
  12. Downloaded the beta on my PowerBook G4 and so far it works great, thank you!

    It scores 447 points in HTML5 Test, better than any other browser. I even did a small comparison with TFF 17 and 31:

    TFF version: 38, 31 17
    V8 Benchmark: 1415, 189, 605 (bigger - better)
    ZDnet load time: 37, 40, 60 (smaller - better)

    Great work!

    ReplyDelete
  13. The new Instagram design also suffers from the Ion/JS Facebook bug.

    STR: Go to my account

    https://instagram.com/chris_hitarori/

    scroll down & click "Load more". It does load more. So far, so good. Then scroll down further. When you reach the bottom, it should load even more thumbnails automatically, but doesn't most of the time. If I switch off Ion, it does, always.

    You don't need to be logged in to Instagram to test this. I know Instagram is sort of the same as Facebook, but maybe here it's easier to debug.

    ReplyDelete
    Replies
    1. It wasn't unfortunately easier to debug, but it was helpful to use to show that the fix I have in hand works on both sites. :)

      Delete
    2. Okay :-) I'm looking forward to test it. Actually I'm surprised there's been only this one major problem with the new JS engine. I've been using 38 exclusively these weeks and haven't found a single new bug other than the ones we already know.

      Delete
  14. I found a bug in 38.0.1, where all popups (hamburger menu, edit the bookmark, remember the password) are not clickable when TFF is on the secondary screen on the left of my main screen.

    ReplyDelete
    Replies
    1. I don't have a two monitor setup, so I can't do much with this report. If you find the cause, let me know.

      Delete
  15. What is more, I don't see checkmarks next to context menu items, e.g. when right clicking on a Console tab in developer tools.

    ReplyDelete
    Replies
    1. OK, I can reproduce this. This is a lower priority issue because they still work (just without the checkmarks), so it's not something I'm going to delay the release for, but I acknowledge the bug and we'll see if I can fix it in either the final or the release after that.

      Delete
  16. Localization update: 6/7 installers done. Still missing: Polish.

    By the way, we *could* have Swedish and Asturian langpacks, too, very easily. We used to have them for TFF 17 and only need these two additional lines translated:

    "…and all our localizers" [for Thank Yous in the About dialog]
    "TenFourFox does not support plugins for this content."

    ReplyDelete
    Replies
    1. I'm Danish som my Swedish might not be the best, but here goes:
      "…och alla våra översättare"
      "TenFourFox inte stöder plugins för detta innehåll."

      Delete
    2. Thank you. I would actually prefer to have this checked by a native speaker of Swedish if possible. Anyone?

      Delete

Due to an increased frequency of spam, comments are now subject to moderation.