180306 - Global History doesn't give up trying to open its db if it fails and it doesn't work unless it can open its db

Status: RESOLVED INVALID

(Core Graveyard :: History: Global, defect)

Description

[timeless]

I'm running with all sorts of interesting patches, i think the most interesting
one leaves me without a profile directory which in turn should result in things
that want a profile directory (like global history) being unable to create files.

This isn't the normal way for things to fail, but I can describe some more
typical cases where mozilla (or an embedder) might not be able to open/write to
global history.  Running mozilla from a CDRom or in a kiosk are two examples. 
It's also possible that someone might need to change the mount settings on their
home drive to readonly for some period. If they do that, they might remount rw
before quiting mozilla, in which case it'd be nice if we wrote the file out at
the end of the session.

WARNING: NS_ENSURE_TRUE(NS_SUCCEEDED(rv)) failed, file
i:/build/mozilla/xpfe/components/history/src/nsGlobalHistory.cpp, line 2368
WARNING: NS_ENSURE_TRUE(NS_SUCCEEDED(OpenDB())) failed, file
i:/build/mozilla/xpfe/components/history/src/nsGlobalHistory.cpp, line 612
WARNING: NS_ENSURE_TRUE(NS_SUCCEEDED(mGlobalHistory->AddPage(spec.get())))
failed, file i:/build/mozilla/docshell/base/nsDocShell.cpp, line 6410

I count 48 attempted opens just for my start page http://www.mozilla.org/start/
Afaik, there's one html page, one stylesheet, one icon, one banner, and the
"download" image because i haven't set BUILD_OFFICIAL recently. -- ~10 opens per
object seems extreme.

Actually trying to go somewhere (home, probably the same page), i get more
WARNING: NS_ENSURE_TRUE(NS_SUCCEEDED(OpenDB())) failed, file
i:/build/mozilla/xpfe/components/history/src/nsGlobalHistory.cpp, line 1182
WARNING:
NS_ENSURE_TRUE(NS_SUCCEEDED(mGlobalHistory->IsVisited(resolvedPath.get(),
&isVisited))) failed, file i:/build/mozilla/docshell/base/nsWebShell.cpp, line 776

And best of all, i don't get any history in the global history window.

Issues:
1. history should work without a backing store.
2. history should probably give up trying to create a backing store and remember
that it failed for the duration of a profile's session.
3. history can attempt to flush at session-logout.

Comment 1

[Alec Flett]

While I think these are great ideas, I think the reality is that our current
history implementation is not going to work without a backing store. (I suspect
changes to Mork would be required for that..and I'd love to see that happen, but
I think thats a much harder problem)

What should be happening is that history should just quietly ignore failure if
it failed to create a history DB the first time. The assertions are probably a
little more aggressive than they need to be, I'm sure an NS_WARNING would
suffice (my fault, perhaps overuse of NS_ENSURE_*)

Comment 2

[:Gavin Sharp [email: [e-mail]]]

No one uses this code anymore, AFAIK (seamonkey and Firefox have moved to places).