In bug 754857 we stopped to expose readonly state on text leafs what causes certain problems with dialogs in NVDA, see 754857 comment #15.

so both entry and text leaf accessibles are exposed as ROLE_SYSTEM_TEXT role on MSAA. Entries can expose editable (IA2) or readonly (MSAA) state. If it's pure MSAA client then there's no way to distinguish whether this accessible is editable entry or a text leaf. The interesting thing is IA2 clients don't need text leafs in general (except it seems the Jamie's case) since they operates on text interface. We could expose static text for all text leafs (which is not consistent to IE though) as Jamie noticed. But I'm not sure if this role is suitable for text leafs inside editable area (like midas document). As long as we need to support MSAA clients then we need to fix it (I hope Marco can dig into problem deeply and see if we have other regressions). Otherwise we can stop to expose text leafs on windows at all (we do that on atk now).

As noted in bug 754857 comment 13, if we're the only ones affected by this, I can change this in NVDA. It'll already be broken for a few Firefox versions anyway, so if we fix it, those versions will work for our users. The question is whether this is the right decision and whether it breaks anyone else. (In reply to alexander :surkov from comment #1) > so both entry and text leaf accessibles are exposed as ROLE_SYSTEM_TEXT role > on MSAA. Entries can expose editable (IA2) or readonly (MSAA) state. If it's > pure MSAA client then there's no way to distinguish whether this accessible > is editable entry or a text leaf. Actually, we can check for the focusable state (and not unavailable). This is what I will do if I need to fix this in NVDA. > The interesting thing is IA2 clients don't > need text leafs in general (except it seems the Jamie's case) since they > operates on text interface. Interestingly, object navigation is much nicer with text leaves as well. The way Gecko does IAText works well enough for traversal, but not so well for navigation. I'm curious as to what you do on Mac (since Mac doesn't do the embedded object thing either), but that's probably off-topic. :) > We could expose static text for all text leafs (which is not consistent to > IE though) as Jamie noticed. But I'm not sure if this role is suitable for > text leafs inside editable area (like midas document). Tricky. They're not themselves editable, though, so I think it's okay. > Otherwise we can stop to expose text leafs on windows at all (we do that on > atk now). See above re object navigation, though I'll grant most users don't do this for Firefox. However, I *think* we need text leaves for our mouse tracking code as well. Mick, am I correct here?

I've implemented specific code to handle Mozilla text leaves in NVDA (regardless of version), so this bug isn't important for us any more. Related NVDA issue ticket: http://www.nvda-project.org/ticket/2613

(In reply to James Teh [:Jamie] from comment #2) > As noted in bug 754857 comment 13, if we're the only ones affected by this, > I can change this in NVDA. It'll already be broken for a few Firefox > versions anyway, so if we fix it, those versions will work for our users. > The question is whether this is the right decision and whether it breaks > anyone else. Marco didn't reported back so I assume we are in good shape. > (In reply to alexander :surkov from comment #1) > I'm curious as to what you do on Mac (since Mac doesn't do the > embedded object thing either), but that's probably off-topic. :) I think they navigate by text leafs. > > We could expose static text for all text leafs (which is not consistent to > > IE though) as Jamie noticed. But I'm not sure if this role is suitable for > > text leafs inside editable area (like midas document). > Tricky. They're not themselves editable, though, so I think it's okay. we used to expose static text accesibles as static text attribute. Maybe it's name collision though. > > Otherwise we can stop to expose text leafs on windows at all (we do that on > > atk now). > See above re object navigation, though I'll grant most users don't do this > for Firefox. However, I *think* we need text leaves for our mouse tracking > code as well. I see. (In reply to James Teh [:Jamie] from comment #3) > I've implemented specific code to handle Mozilla text leaves in NVDA > (regardless of version), so this bug isn't important for us any more. cool, thank you

wontfix then?