On Thu, Sep 4, 2008 at 9:00 PM, Hal Duston <hald@kc.rr.com> wrote:

> That isn't possible unless you want to do an exec call every time the user
> goes to a new page. That's why Chrome does its own task manager that knows
> what each thread is doing.

Postgresql and sendmail manage it.

root 5082 1 0 Jul27 ? 00:52:53 sendmail: accepting connections
smmsp 5091 1 0 Jul27 ? 00:00:12 sendmail: Queue runner@01:00:00 for /var/spool/clientmqueue
postgres 16733 16667 0 Aug04 ? 00:01:32 postgres: logger process
postgres 16737 16667 0 Aug04 ? 00:06:53 postgres: writer process
postgres 16738 16667 0 Aug04 ? 00:05:28 postgres: wal writer process
postgres 16739 16667 0 Aug04 ? 00:03:30 postgres: autovacuum launcher process
postgres 16740 16667 0 Aug04 ? 00:05:07 postgres: stats collector process

They both manage that by recreating the environment of the already running process.

Yes, they exec to spawn these separate processes. But those processes can't change the arguments under which they were invoked without themselves exec-ing all over again. So when you click on a "Heh.", you'd have to change one process from
monster 3952 26665 0 10:08 ? /usr/lib/firefox-4.0/firefox-bin -UILocale en-US instapundit.com
to
monster 3952 26665 0 10:08 ? /usr/lib/firefox-4.0/firefox-bin -UILocale en-US johndoe.blogspot.com/2008/09/snarky-article-title-here.html

The only way I know to do that is to exec firefox-bin, invoking the new URL as an argument. What I don't know is how much overhead that takes. I suppose a stripped-down binary, that inherits a pre-built environment populated by another binary reading configs, etc., might not be doable.

But that's not the way that any existing browser I know of is written. You're talking about a complete re-factoring of how browsers work. That's not to say that it's a bad idea, just that it's a non-trivial effort, which might make things much slower