Monday, March 29, 2021

The end of TenFourFox and what I've learned from it

Now that I have your attention.

I've been mulling TenFourFox's future for awhile now in light of certain feature needs that are far bigger than a single primary developer can reasonably embark upon, and recent unexpected changes to my employment, plus other demands on my time, have unfortunately accelerated this decision.

TenFourFox FPR32 will be the last official feature parity release of TenFourFox. (A beta will come out this week, stay tuned.) However, there are still many users of TenFourFox — the update server reports about 2,000 daily checkins on average — and while nothing has ever been owed or promised I also appreciate that many people depend on it, so there will be a formal transition period. After FPR32 is released TenFourFox will drop to security parity and the TenFourFox site will become a placeholder. Security parity means that the browser will only receive security updates plus certain critical fixes (as I define them, such as crash wallpaper, basic adblock and the font blacklist). I will guarantee security and stability patches through and including Firefox 93 (scheduled for September 7) to the best of my ability, which is also the point at which Firefox 78ESR will stop support, and I will continue to produce, generate and announce builds of TenFourFox with those security updates on the regular release schedule with chemspills as required. There will be no planned beta releases after FPR32 but Tenderapp will remain available to triage bugfixes for new changes only.

After that date, for my own use I will still make security patches backported from the new Firefox 91ESR publicly available on Github and possibly add any new features I personally need, but I won't promise these on any particular timeline, I won't make or release any builds for people to download, I won't guarantee any specific feature or fix, I won't guarantee timeliness or functionality, and there will be no more user support of any kind including on Tenderapp. I'll call this "hobby mode," because the browser will be a hobby I purely maintain for myself, with no concessions, no version tags (rolling release only), no beta test period and no regular schedule. You can still use it, but if you want to do so, you will be responsible for building the browser yourself and this gives you a few months to learn how. Also, effective immediately, there will be no further updates to TenFourFoxBox, the QuickTime Enabler, the MP4 Enabler or the TenFourFox Downloader, though you will still be able to download them.

Unless you have a patch or pull request or it's something I care about, if you open an issue on Github it will be immediately closed. Similarly, any currently open issues I don't intend to address will be wound down over the next few weeks. However, this blog and the Github wiki will still remain available indefinitely, including all the articles, and all downloads on SourceForge will remain accessible as well. I'll still post here as updates are available along with my usual occasional topics of relevance to Power Mac users.

Classilla, for its part, is entering "hobby mode" today and I will do no further official public work on it. However, I am releasing the work I've already done on 9.3.4, such as it is, plus support for using Crypto Ancienne for self-hosted TLS 1.2 if you are a Power MachTen user (or running it in Classic or under Mac OS in Rhapsody). You can read more about that on Old VCR, my companion retrocomputing blog.

I'm proud of what we've accomplished. While TenFourFox was first and foremost a browser for me personally, it obviously benefited others. It kept computers largely useable that today are over fifteen years old and many of them even older. In periods of a down economy and a global pandemic this helped people make ends meet and keep using what they had an investment in. One of my favourite reports was from a missionary in Myanmar using a beat-up G4 mini over a dialup modem; I hope he is safe during the present unrest.

I'm also proud of the fair number of TenFourFox features that were successfully backported or completely new. TenFourFox was the first and still one of the few browsers on PowerPC Mac OS X to support TLS 1.3 (or even 1.2), and we are the only such browser with a JavaScript JIT. We also finished a couple features long planned for mainline Firefox but that never made it, such as our AppleScript (and AppleScript-JavaScript bridge) support. Our implementation even lets you manipulate webpages that may not work properly to function usefully. Over the decade TenFourFox has existed we also implemented our own native date and time controls, basic ad block, advanced Reader View (including sticky and automatic features), additional media support (MP3, MP4 and WebP), additional features and syntax to JavaScript, and AltiVec acceleration in whatever various parts of the browser we could. There are also innumerable backported bug fixes throughout major portions of the browser which repair long-standing issues. All of this kept Firefox 45, our optimal platform base, useful for far longer than the sell-by date and made it an important upstream source for other legacy browsers (including, incredibly, OS/2). You can read about the technical differences in more detail.

Many people have contributed to TenFourFox and to the work above, and they're credited in the About window. Some, like Chris T, Ken Cunningham and OlgaTPark, still contribute. I've appreciated everyone's work on the source code, the localizations and their service in the user support forums. They've made the job a little easier. There are not enough thank yous for these good people.

When September rolls around, if you don't want to build the browser yourself it is possible some downstream builders like Olga may continue backports. I don't speak for them and I can't make promises on their behalf. Olga's builds run on 10.4, 10.5 and 10.6. If you choose to make your own builds and release them to users, please use a different name for your builds than TenFourFox so that I don't get bothered for support for your work (Olga has a particular arrangement with me but I don't intend to repeat it for others).

You might also consider another browser. On PowerPC 10.5 your best alternative is Leopard WebKit. It has not received recent updates but many of you use it already. I don't maintain or work on LWK, but there is some TenFourFox code in it, and Tobias has contributed to TenFourFox as well. If you don't want to use Safari specifically, LWK can be relinked against most WebKit shells including Stainless and Roccat.

If you are using TenFourFox on 10.6, you could try using Firefox Legacy, which is based on Firefox 52. It hasn't been updated in about a year but it does have a more recent platform base than official Firefox for 10.6 or TenFourFox.

However, if you are using TenFourFox on 10.4 (PowerPC or Intel), I don't have any alternative suggestions for you. I am not aware of any other vaguely modern browser that supports Tiger. Although some users have tried TenFourKit, it does not support TLS 1.1 or 1.2 (only Opera 10.63 does), and OmniWeb, Camino, Firefox 3.6 and the briefly available Tor Browser for PowerPC are now too old to recommend for any reasonable current use.

So, that's the how. Here's the why and what. I have a fairly firm rule that I don't maintain software I don't personally use. The reason for that is mostly time, since I don't have enough spare cycles to work on stuff that doesn't benefit me personally, but it's also quality: I can't maintain a quality product if I don't dogfood it myself. And my G5 has not been my daily driver for a good couple years; my daily driver is the Raptor Talos II. I do use the G5 but for certain specific purposes and not on a regular daily basis.

Additionally, I'm tired. It's long evenings coding to begin with, but actual development time is only the start of it. It's also tying up the G5 for hours to chug out the four architecture builds and debug (at least) twice a release cycle, replying to bug reports, scanning Bugzilla, reading the changelogs for security updates and keeping up with new web features in my shrinking spare time after doing the 40+-hour a week job I actually got paid for. Time, I might add, which is taken away from my other hobbies and my personal relaxation, and time which I would not need to spend if I did this purely as a hobby and never released any of it. Now that Firefox is on a four-week release schedule, it's just more than I feel I can continue to commit to and I'm neglecting the work I need to do on the system that I really do use every day.

We're running on fumes technologically as well. Besides various layout and DOM features we don't support well like CSS grid, there are large JavaScript updates we'll increasingly need which are formidably complex tasks. The biggest is async and await support which landed in Firefox 52, and which many sites now expect to run at all. However, at the time it required substantial changes to both JavaScript and the runtime environment and had lots of regressions and bugs to pick up. We have some minimal syntactic support for the feature but it covers only the simplest of use cases incompletely. There are also front end changes required to deal with certain minifiers (more about this in a moment) but they can all be traced back to a monstrous 2.5MB commit which is impossible to split up piecemeal. We could try to port 52ESR as a whole, but we would potentially suffer some significant regressions in the process, and because there is no Rust support for 32-bit PowerPC on OS X we couldn't build anything past Firefox 54 anyway. All it does is just get us that much closer to an impenetrable dead end. It pains me to say so, but it's just not worth it, especially if I, the browser's only official beneficiary, am rarely using it personally these days. It's best to hang it up here while the browser still works for most practical purposes and people can figure out their next move, rather than vainly struggling on with token changes until the core is totally useless.

Here is what I have learned working on TenFourFox and, for that matter, Classilla.

Writing and maintaining a browser engine is fricking hard and everything moves far too quickly for a single developer now. However, JavaScript is what probably killed TenFourFox quickest. For better or for worse, web browsers' primary role is no longer to view documents; it is to view applications that, by sheer coincidence, sometimes resemble documents. You can make workarounds to gracefully degrade where we have missing HTML or DOM features, but JavaScript is pretty much run or don't, and more and more sites just plain collapse if any portion of it doesn't. Nowadays front ends have become impossible to debug by outsiders and the liberties taken by JavaScript minifiers are demonstrably not portable. No one cares because it works okay on the subset of browsers they want to support, but someone bringing up the rear like we are has no chance because you can't look at the source map and no one on the dev side has interest in or time for helping out the little guy. Making test cases from minified JavaScript is an exercise in untangling spaghetti that has welded itself together with superglue all over your chest hair, worsened by the fact that stepping through JavaScript on geriatic hardware with a million event handlers like waiting mousetraps is absolute agony. With that in mind, who's surprised there are fewer and fewer minority browser engines? Are you shocked that attempts like NetSurf, despite its best intentions and my undying affection for it, are really just toys if they lack full script runtimes? Trying and failing to keep up with the scripting treadmill is what makes them infeasible to use. If you're a front-end engineer and you throw in a dependency on Sexy Framework just because you can, don't complain when you only have a minority of browser choices because you're a big part of the problem.

Infrastructure is at least as important as the software itself. A popular product incurs actual monetary costs to service it. It costs me about US$600 a month, on average, to run my home data center where Floodgap sits (about ten feet away from this chair) between network, electricity and cooling costs. TenFourFox is probably about half its traffic, so offloading what we can really reduces the financial burden, along with the trivial amount of ad revenue which basically only pays for the domain names. Tenderapp for user support, SourceForge for binary hosting, Github for project management and Blogger for bloviating are all free, along with Google Code where we originally started, which helped a great deal in making the project more sustainable for me personally even if ultimately I was shifting those ongoing costs to someone else. However, the biggest investment is time: trying to stick to a regular schedule when the ground is shifting under your feet is a big chunk out of my off hours, and given that my regular profession is highly specialized and has little to do with computing, you can't really pay me enough to dedicate my daily existence to TenFourFox or any other open-source project because I just don't scale. (We never accepted donations anyway, largely to avoid people thinking they were "buying" something.) I know some people make their entire living from free open source projects. I think those people are exceptions and noteworthy precisely because of their rarity. Most open source projects, even ones with large userbases, are black holes ultimately and always will be.

Gecko has a lot of technical baggage, but it is improving by leaps and bounds, and it is supported by an organization that has the Internet's best interests at heart. I have had an active Bugzilla account since 2004 and over those 16+ years I doubt I would have gotten the level of assistance or cooperation from anyone else that I've received from Mozilla employees and other volunteers. This is not to say that Mozilla (both MoFo and MoCo) has not made their blunders, or that I have agreed personally with everything they've done, and with respect to sustainability MoCo's revenues in particular are insufficiently diversified (speaking of black holes). But given my experience with other Mozillians and our shared values I would rather trust Mozilla any day with privacy and Web stewardship than, say, Apple, who understandably are only interested in what sells iDevices, and Google, who understandably are only interested in what improves the value proposition of their advertising platforms. And because Chrome and Chromium effectively represent the vast majority of desktop market share, Google can unilaterally drive standards and force everyone to follow. Monopolies, even natural ones, may be efficient but that doesn't make them benign. I'll always be a Firefox user for that reason and I still intend to continue contributing code.

Now for the mildly controversial part of this post and the one that will make a few people mad, but this is the end of TenFourFox, and a post-mortem must be comprehensive. For this reason I've chosen to disable comments on this entry. Here is what you should have learned from TenFourFox (much the same thing users should have learned from any open-source project where the maintainer eventually concluded it was more trouble than it was worth).

If you aren't paying for the software, then please don't be a jerk. There is a human at the other end of those complaints and unless you have a support contract, that person owes you exactly nothing. Whining is exhausting to read and "doesn't work" reports are unavoidably depressing, disparaging or jokey comments are unkind, and making reports nastier or more insistent doesn't make your request more important. This is true whether or not your request is reasonable or achievable, but it's certainly more so when it isn't.

As kindly as I can put it, not all bug reports are welcome. Many are legitimately helpful and improve the quality of the browser, and I did appreciate the majority of the reports I got, but even helpful bug reports objectively mean more work for me though it was work I usually didn't mind doing. Unfortunately, the ones that are unhelpful are at best annoying (and at worst incredibly frustrating) because they mean unhappy people with problems that may never be solvable.

The bug reports I liked least were the ones that complained about some pervasive, completely disabling flaw permeating the entire browser from top to bottom. Invariably this was that the browser "was slow," but startup crashes were probably a distant second place. The bug report would inevitably add something along the lines of this should be obvious, or talk about the symptom(s) as if everyone must be experiencing it (them).

I'm not doubting what people say they're seeing. But you should also consider that asserting the software has such a grave fault effectively alleges I either don't use the software or care about it, or I would have noticed. Most of the time my reply was to point out that my reply was being made in the browser itself, and to point out that we had regular beta phases where the alleged issue had not surfaced, so no, it must not be that pervasive, and let's figure out why your computer behaves that way. As far as the browser being slow, well, that's part personal expectation and part technical differences. TenFourFox would regularly win benchmarks against other Power Mac browsers because its JavaScript JIT would stomp everything else, but its older Mozilla branch has weaker pixelpushing and DOM that is demonstrably slower than WebKit, and no Power Mac browser is going to approach the performance you would get on an Intel Mac with any browser. Some of this is legitimate criticism, but overall if that's what you're expecting, TenFourFox will disappoint you. And it certainly did disappoint some people, who felt completely empowered to ignore all that context and say so.

Here is another unwelcome bug report, sometimes part of those same reports: "Version X+1 does something bad that Version X didn't, so I went back to Version X (or I've switched to another browser). Please let me know when it's fixed."

As a practical consideration, if you have such a serious issue where you can't use the browser for your desired purpose then I guess you do what you gotta do. But consider you may also be saying that you don't care about solving the problem. Part of it is, like the last report, making the sometimes incorrect assumption that everyone else must be seeing what you're seeing. But the other part is because you've already reverted to the previous release, you don't have any actual investment in the problem being solved. If it actually is a problem that can be fixed, and I do fix it, you're using the previous version and may or may not be in a position to test it. But if it's actually a problem I can't observe, then it won't get fixed assuming it actually does exist, because I don't see that problem on Version X+1 myself and the person who can see it, i.e., you, has bailed out. If you want me to fix it, especially if you are unwilling or unable to fix it yourself, then you need to stick with it like I'm sticking with it.

What should you do? Phrase it better. Post your reports with the attitude that you are just one user, using free software, from the humility of your own personal experience on your own system. Make it clear you don't expect anything from the report, you are grateful the software exists, you intend to keep using it and this is your small way of giving back. Say this in words because I can't see your face or hear your voice. Write "thank you" and mean it. Acknowledge the costs in time and money to bring it to you. Tell me what's good about it and what you use it for. That's how you create a relationship where I can see you as a person and not a demand request, and where you can see me as a maintainer and not a vending machine. Value my work so that I can value your insights into it. Politeness, courtesy and understanding didn't go out the window just because we're interacting through a computer screen.

Goodness knows I haven't been perfect and I've lost my temper at times with people (largely justifiably, I think, but still). All of us are only human. But today, looking back on everything that's happened, I'm still proud of TenFourFox and I'm still glad I started working on it over 10 years ago. Here's the first functional build of Firefox 4.0b7pre on Tiger (what became the first beta of TenFourFox), dated October 15, 2010:

This was back when Mozilla was sending thank-you cards to Firefox 4 beta testers:
TenFourFox survived a lot of times when I thought it was finished for one technical reason or another, and it's still good enough for the couple thousand people who use it every day and the few thousand more who use it occasionally. It kept a lot of perfectly good hardware out of landfills. And most of all, it got me years more out of my own Quad G5 and iBook G4 and it still works well enough for the times I do still need it. Would I embark upon it again, knowing everything I know now and all the work and sweat that went into it? Heck yeah. In a heartbeat.

It was worth it.

Friday, March 19, 2021

TenFourFox FPR31 available

TenFourFox Feature Parity Release 31 final is now available for testing (downloads, hashes, release notes). There are no additional changes from the beta except for outstanding security patches. Locale langpacks will accompany this release and should be available simultaneously on or about Monday or Tuesday (March 22 or 23) parallel to mainline Firefox.

Friday, March 12, 2021

TenFourFox FPR31b1 available (now with site-specific user agent UI and Auto Reader View)

TenFourFox Feature Parity Release 31 beta 1 is now available (downloads, hashes, release notes). I didn't get everything done here that I wanted to, though thanks to Chris T I do have a reproducing local test version of the infamous issue 621; at least I am able now to see that it's clearly a problem in the JavaScript parser generating something incorrectly, but I'm still not able to tell where the specific deficiency lies.

However, there's still new stuff in this release. Olga T Park contributed a backport from later Firefox versions to fix saving passwords in private browsing, and I also finished fully exposing support for site specific user agents. This was quietly reimplemented in FPR17 for interested users, but now that it's getting more and more necessary on more and more sites, I have made the feature a visible and supported part of the browser interface. Instead of having to enter sites and strings manually into about:config, though you still can, you can now go to the TenFourFox preference pane,

and click the new "Site Specific" button under User Agent. A new dialogue box will open.
Site-specific user agents in use appear in the bottom half of the window with domain and user agent strings. You can enter anything you want for a user agent string, as shown, or you can pick a pre-defined one to fill the box from the dropdown (the same ones you would be offered for the global user agent option, which remains supported as well). In this example I've chosen a random "what's my user agent" domain and assigned the Classilla string to it. We click Save Changes and try it out:
Ta-daa. As before, having site-specific user agent strings does slightly slow the browser down, though most of the penalty is paid on the first and not so much on additional strings you enter. Our implementation has no penalty if you have no site-specific user agents loaded and this remains the default. For sites that are unspecified, obviously the global user agent option still applies. If you have a user-agent add-on already installed, you can still use it, but it may have interactions if you try to use this feature at the same time and you're on your own if you do.

I was also planning to do a Reader View update for this release, which will need a user interface of its own, but every new UI feature requires additional locale strings and I wanted to give our localizers (led by Chris) a chance to catch up on the new strings in time for the final release on March 22. However, I do have a not-yet-exposed feature's plumbing done, which is another enhancement to Reader View: auto Reader View.

Auto Reader View is different from sticky Reader View, which has been the default since FPR27. Sticky Reader View means that when you go into Reader View, links you click on also load in Reader View, until you quit it by clicking one of the exit buttons. Auto Reader View, however, allows you to tell the browser to automatically open pages from a domain in Reader View as soon as you click on any link to that domain from any page, in Reader View or not. Since front pages may not work as well, you can specify to do this just for "subpages" or for all pages.

An example is the Los Angeles Times, which is more or less my semi-local newspaper here in Southern California. Its article pages have a lot of irritating popup divs, so go into about:config, create a new string preference called tenfourfox.reader.auto.www.latimes.com and set it to the single letter s (for subpages; for all pages, use y for, um, "yes"). Now, visit the L.A. Times front page. It renders in the normal browser view, but if you click on any article, it immediately shifts to Reader View. Click the back button and you're back on the front page. If you decide you do want to see the article as intended, just click the Reader View icon in the address bar as usual, and the article will render in "full" form; links to subpages you click on from there will go back to Reader View.

As I've said many times, I think Reader View is an important way of making sites render faster and more usefully, especially on G3 and low-end G4 systems. It also cuts out a lot of crap, meaning it's back to only the columnists annoying me in the L.A. Times and not the popups. For FPR32 I will be updating the internal Reader View to the current version of Readability.js, adding a preference pane UI similar to site-specific user agents, and maybe also adding the current Firefox feature allowing you to adjust the gutter margins wider or narrower. But for now you can experiment with the feature and let me know how functional or useful it is to you. FPR31 final comes out on or around March 22, parallel with Firefox 78.9 and 87.