linux: work around gnome keyring memory corruption

linux: work around gnome keyring memory corruption

Older versions of gnome keyring corrupt memory if you fetch too much
data from it.  Despite our best efforts, we couldn't come up with a way
to tell if you're using the broken old version so that we could blacklist
it.  Instead, we fetch passwords one-by-one, which is slow but tolerable
since we do it on a background thread anyway.

While I'm here, refactor the code a bit to simplify it.

TEST=password manager doesn't crash when using mdm's sample 100 password db
with --password-store=gnome on hardy

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=50771

[Mike Mammarella, 2010-06-24 05:11:09]

Thanks! LGTM with a few comments.

http://codereview.chromium.org/2838023/diff/2001/3001
File chrome/browser/password_manager/native_backend_gnome_x.cc (right):

http://codereview.chromium.org/2838023/diff/2001/3001#newcode54
chrome/browser/password_manager/native_backend_gnome_x.cc:54:
typeof(&gnome_keyring_##name) wrap_gnome_keyring_##name;
Since you declare our function pointers with types based on the gnome keyring
prototypes, we don't need VerifyGnomeKeyringTypes below any more. Good idea!

http://codereview.chromium.org/2838023/diff/2001/3001#newcode63
chrome/browser/password_manager/native_backend_gnome_x.cc:63:
wrap_gnome_keyring_##name = gnome_keyring_##name;
name = name was to avoid unused variable warnings, but if you remove this anyway
then it doesn't matter.

http://codereview.chromium.org/2838023/diff/2001/3001#newcode139
chrome/browser/password_manager/native_backend_gnome_x.cc:139: // We don't do
the hacky version check here. When linking directly, we assume
This comment can go away.

http://codereview.chromium.org/2838023/diff/2001/3001#newcode453
chrome/browser/password_manager/native_backend_gnome_x.cc:453: // reasonable
unofficial way to do it.  So we workaround by using a
Nit: the verb form of workaround is two words.