Przejdź do treści

Pieśni o Bezpieczeństwie Sieci
blog Michała "ryśka" Woźniaka

Long-expected KMail2 rant

To jest bardzo stary wpis, opublikowany ponad 4 lata temu.
Możliwe, że nie odzwierciedla dziś poglądów Autora lub zewnetrznych faktów. Jest zachowany jako wpis historyczny.
Niestety, polskie tłumaczenie nie jest dostepne. Wyświetlono w języku: English.

WARNING: This is a rant. You have been warned.

This bragpost has been growing on me for some time now. About since the very day I had the questionable pleasure of trying out Akonadi-based KMail2. That would be about 2 years, and back then KMail2 was buggy and slow.

Today it is buggy and on the verge of being almost possible to consider as trying to be usable. That’s how far this project has gone in these 2 years.

Good Ole Days

In the Good Ole Days, and by that I mean “pre-KMail2”, there were more or less two kinds of FLOSS e-mail clients out there.

Bad UI, decent engineThunderbird was an example here; compared to KMail 1.x, Thunderbird had completely absurd layout of configuration options and doing simple things (like creating a new e-mail identity and connecting it to SMTP and IMAP/POP accounts) seemed daunting.

Completely subjectively I include in this group also Evolution and Mutt, simply because I use KDE (hence GTK is bad mojo) and am a GUI afficionado (yeah, yeah, so shoot me).

Great UI, great engine – in my book, this group consisted of just one project: KMail 1.x. It had everything: IMAP/POP/SMTP, TLS/SSL/STARTTLS, Sieve, advanced local filters, multiple identities connectable not only to specific accounts but to specific mail folders, proper standards support, proper quoting, and everything configurable via a sane interface with predictable results. All mail kept nicely in standard Maildirs so that KMail’s mail was the easiest to handle by any standard Maildir tool.

And from verions to version there actually were incremental improvements still. What’s not to love!

What the hell happened?

Since then the KMail2 team apparently decided to take a different route, which I would guess was something along the lines of:

  1. take the best e-mail client out there;
  2. mangle in ways nobody would believe were possible;
  3. ???
  4. PROFIT!!

And thus today KMail2 lands squarely in the “WTF UI, WTF engine” category (at least retaining the “the only one in that category” title).

So, what’s wrong with KMail2, exactly?

Well, what isn’t?.. Not really even sure where to start. So let’s start with the “migration assistant”.

Migration assistant

The first time a user runs KMail2 over an existing KMail 1.x setup, a “migration assistant” appears. Not much to it, it just informs the user about the need for migration and then attempts to migrate all the accounts and e-mails from KMail 1.x. this takes a lot of time, usually, but hey, that’s the price of progress, right? Once the migration is finished the user is set up and are ready to go.

What the user doesn’t know, however, is that the end of the migration is just the beginning of fun times. During the next few days the user will notice many entertaining facts.

Like the fact that the migration assistant does not migrate e-mail filters. Got 200+ KMail 1.x filters set-up to help you stay afloat on the flood of everydays e-mail activity? Tough luck, buddy, that’s history. Good luck re-creating them all.

Like missing e-mails (or duplicate e-mails). Or that viewing e-mails that actually got migrated will be unbearably slow at times. Some folders will just get marked as “corrupted” (yes, these folders that were created by the “migration assistant” itself, why do you ask?) and will cause KMail2 to block the whole account entirely (with a nice red tint on all folder names) until KMail2 and Akonadi get restarted…

As it turns out you can actually skip the migration and import messages from KMail 1.x store later on, and that seems just a bit better idea (caveat: “better” does not mean “perfect”, or even “good”).

But wait, why the heck do I need to migrate my 10GiB of e-mail history from a perfectly working (and standard!) set-up of MailDirs at all, just to use KMail2? And what the heck is Akonadi?


Akonadi is supposed to be the data engine behind all PIM (Personal Information Manager) related content in KDE. It handles contacts, calendars, e-mail and possibly other data (the list is growing). And it was released as stable as soon as it achieved more-or-less alpha status. At least that’s what I can tell from how well it does the job at hand.

The fun part with Akonadi is that it has to have a database back-end set-up. By default it uses MySQL, so when you have Akonadi, you have MySQL instance running.

But hey, relational databases are a true and tested technology, why not use it, instead of some internal, in-memory data store that Akonadi would use either way? And of course now Akonadi doesn’t have to use any internal, in-memory data store for data retrieved from such a database. Surely, the additional abstraction layer won’t change much, at least not to a point where it’s evident Akonadi is a problem.

Finally, it’s not as if keeping everything in nice, editable, standards-compliant files on disk makes anything (backups? trying a different e-mail client? export/import? data recovery after a failure?) easier and safer, right?..


Some problems with KMail2 are related to Akonadi; some are clearly the fault of KMail2 itself. Of both kinds there are many.

The first biggie is the slowness. KMail 1.x displayed the list of e-mails in a given folder instantly. And I mean instantly. Same thing with displaying the contents of an e-mail once clicked. In KMail2 I have to wait several seconds for the list to appear and be usable (i.e. the list can be visible, but not clickable). That is probably related to the fact that e-mail data are now retrieved from Akonadi, but why do I care? From the end-user perspective this is worse.

Now, displaying the folder contents is one thing. Moving a folder with a few hundred or thousand e-mails from one place in a folder tree to another – that’s a whole different game. This can take about a minute (yes, I have actually timed it) before it’s done and the interface is usable again.

Oh, and you get no visual cues that that’s the case – the folder list looks perfectly normal. You can try to click a folder, only you’ll not see the contents. You’ll also notice that your CPU fan goes berserk and the whole interface grinds almost to a halt… and then magically, the folder being moved appears in the place it’s being moved to and the interface is almost usable, again – with no visual cues to that effect.

Then there’s inconsistency. Duplicate e-mails. E-mails that magically become unread again. Folders that have their whole contents disappear after deleting a single out of many e-mails they contain, only to show the missing e-mails after the user starts pondering, frantically, when was the last time a full backup was made…

Like the e-mails in the outbox waiting for user’s explicit “send now”, that get sent automagically once a new Internet connection is available. In the outgoing accounts configuration there even is an option of configuring what should happen with messages in the outgoing folder. This setting is completely ignored. Messages get sent out when KMail2 (or Akonadi?) decides so and that’s that.

Like new e-mail accounts that, upon failure, display random failure messages – sometimes it’s “Authentication failed”, sometimes it’s “KWallet access denied” (even though KWallet was never used, or was explicitly instructed to permit its use and is open), sometimes only a cryptic “Account misconfigured” even though the account worked just minutes ago.

Then there’s mail importing. Supposedly a different, better way to import KMail 1.x e-mails into KMail2 than the “migration assistant”, it does a decent job of importing your e-mails from the old KMail 1.x store and marking them all unread. One would think that if KMail2 is importing e-mails from KMail 1.x store, it would be possible for it to properly import also the status of messages. Apparently, one would be mistaken.

In my case that meant I had 140k (no, not a typo) unread messages I had to sift through and find the few messages that actually were unread. And then mark all the rest as read – which, unsurprisingly, took a lot of time, mainly because even marking a mail folder as unread takes much longer than in KMail 1.x.

And after I’m done with that, there’s the joy of moving the imported folders (conveniently put in a “KMail-Import-mail” folder) to their correct accounts and places in the folder hierarchy. Which, as we already know, is excruciatingly slow.

Finally, the protocol or what-have-you changes. Even though KMail2 and Akonadi are being shipped as stable, the protocol and formats of the internal storage changes just a tiny bit with some minor releases. And that means that you can have all the fun and all the entertainment with “migrating” your e-mails more than once. Indeed, you can be sure you will.

But apart from these important problems, there are scores of smaller annoyances, discovery of which brings endless joy to a happy KMail2 user. I won’t be able to go into detail and describe them all – mainly because there are so many of them. But here are some of my favourites.

Handling of ignored threads – that’s a great feature for mailing lists: the user marks a thread as “ignored” and all future e-mails in that thread get automagically marked “read”. If it only worked! In KMail2 if a thread is marked “ignored” you get 50/50 chance a given new e-mail in it will get marked “ignored” and “read”. You can have (and indeed, I do have) threads marked “ignored” with several unread, un-ignored e-mails inside. Which, of course, defeats the purpose.

But wait, there’s more! Once the user gets annoyed and un-ignores, pretty much all e-mails in it get marked “unread”. Yay!

Password prompt for each and every outgoing e-mail. KMail 1.x had a nice, simple and very usable feature: during sending the password for a given account (if not stored) only had to be supplied once; after all messages have been sent, the password was being forgotten again. So, for 20 outgoing e-mails from a given account the user had to supply the password only once. Usable and safe.

Of course, KMail2 improves upon this idea by asking for the password separately for each and every e-mail.

Which wouldn’t maybe be that annoying had the password prompt been focused properly.

Un-focused password prompt. So, you want to send your e-mail, but the SMTP password is not stored in KWallet? No worries, KMail will display a password prompt for you; but be careful, if start typing your password instantly as soon as the prompt appears, your password will land in the random other application accepting focus, as the e-mail prompt is un-focused (and there seems no way to make it focused by default).

So, instead of:
Ctrl-Enter -> type-in-password -> Enter -> sent
…you get:
Ctrl-Enter -> find-the-damn-prompt-window-and-click-on-it -> type-in-password -> Enter -> sent.

Or, well, you would, if only the following was not the case…

“Send now” means “ask when to send”. In KMail 1.x the “Send now” button and short-cut meant just that: send the message being composed immediately. KMail2 usability experts decided that simple actions like “Send now” are too simple and they need to ask the users what they really want to achieve. Once you click “Send now”, you get a nice dialog box asking you, if you really want to send now, or maybe send layer, or cancel the whole ordeal.

To add insult to injury, there is a setting in outgoing accounts configuration called “Default sending method”, with two options:

  • send immediately;
  • send later.

This option is also completely ignored.

This is completely bollocks. If the user clicks “Send now”, what makes you think, pray tell, that they want to do anything different than, you know, sending the e-mail immediately?.. This adds the need for additional (and completely unnecessary) click for each outgoing message. Usability FAIL.

“Send later” means “ask me in detail”. There is also the “Send later” action. In KMail 1.x it simply saved the composed e-mail directly in the outgoing mail folder, waiting for user to initiate sending.

In KMail2 the user gets an additional dialog box containing a date and time picker, an option to configure repetition of sending of this e-mail (with no explanation if it will be attempted until successful at configured times, or is the same e-mail going to be sent again, and again, and again…), and an option to just “move to outbox”.

While I can see the potential usefulness of an option to configure in detail when a given mail will get sent, this should be a separate option. Users accustomed to the good old perfectly functional “Send later” action are only going to find such a “feature” annoying. That’s another click on every mail they send each day. That’s a lot of unneeded clicks.

End of Rant

I have been a loyal KMail 1.x user for almost a decade. Started using it in KDE 3.x series, I always admired the standards-compliance, the configurability and adaptability, the feature set, the speed. It was so good, indeed, that no other Qt-based clients gained traction. They were unneeded. KMail 1.x simply did the job best.

Compared to KMail 1.x, KMail2 is a sad excuse of a rewrite. It’s all KMail 1.x has never been to me – unstable, buggy, slow, unpredictable, with daunting configuration that has options that get ignored; nigh-unusable with the over-engineered Akonadi under the hood.. This is very sad, as there are no other Qt-based, KDE-integrated, configurable and advanced e-mail clients to choose from.

KMail2 has been in development for years now, and effects are far from satisfactory. Users want to be “saved” from it, bloggers warn about it; the question everybody asks is “Anyone succeeded with kmail2?”.

I am still using KMail2 now, but am on the look-out for some sane Qt-based e-mail client. Trojitá is still too basic for my needs, but at least it’s usable. If you have any suggestions where to look, please do drop a line or comment on Diaspora.