| 2012-05-16 14:04:38 |
Thomas Berezansky |
description |
The initial commit message is reproduced below the link to the branch.
This is a significant enough change that I feel that more than just one person should review it, and discussion in IRC has brought up the thought of "have everyone describe exactly what they tested to ensure we get decent coverage".
As a note, I included the work from the following bugs while building/testing the branch for this one:
Bug #921740 - External sites can be opened in the staff client
This has a component that needed to be re-written to support later XULRunner versions, and would *not* have worked in Extension mode in any satisfactory manner.
Bug #923868 - AccessControl plugin no longer extracts properly
I needed that fix to test clean builds of installers that auto-update.
In addition, this covers Bug #628378 (Remote XUL support going away in Gecko 2) and may help correct Bug #491537 (xulrunner 1.9 (and Firefox 3.5) prints garbage to certain printers) by introducing a significantly later XULRunner version.
In the interest of documenting what has and has not been tested, I touched on the following:
Basic Circ Functions, Basic Admin functions (a number of config menus in particular), Automatic Updates, Installer Building, basic receipt printing, Vandelay loads (but I didn't test much beyond that), and minimal reporter testing. Generally I bounced through random parts of the system I knew enough about to check things in, looking for anything that wasn't working, and tried to hit interfaces I didn't know enough about to ensure that they at least appeared to load without errors. I don't have a comprehensive list of them, though.
http://git.evergreen-ils.org/?p=working/Evergreen.git;a=shortlog;h=refs/heads/collab/tsbere/new_xulrunner
XULRunner/Firefox 4+ change the rules significantly, and to support them
significant changes have to be made in Evergreen.
This commit covers a number of things:
1 - Remote XUL Issues
Later versions of XULRunner disable support for Remote XUL without using a
whitelist entry, and whitelisting may stop working eventually as well. The
solution here is the addition of a new protocol wrapper that allows Remote
XUL to act like Local XUL.
The wrapper is oils:// and acts like HTTPS for the most part. The "host" the
client connects to is always "remote", however, and use of the wrapper has
multiple effects.
The first of these effects is the enabling of Remote XUL content, with as
close to the Local XUL security level as I could manage. This means that
nearly anything Local XUL can do, Remote XUL should be able to do too.
The second of these effects is the disabling of JavaScript-visible cookies.
XUL doesn't have cookies, and as such a side effect of the wrapper allowing
remote content to appear to be local content is the lack of cookies. This
does *not* affect server side visibility of cookies, however, so things like
TPac can still see cookies.
The third of these effects is the "hiding" of the hostname for anything
going through the wrapper. This prevents oils_persist from saving things
correctly. I thus made oils_persist check for the wrapper and use the data
stash to get the hostname instead. I suspect that we can get rid of the
location.hostname check entirely at this point, but have not done so.
The wrapper also sets a request header of OILS-Wrapper, value of 'true'.
That allows the server to know that the request came through the wrapper. I
used that ability to have TPac go into "Staff" mode when the wrapper is in
use. This will enhance extension use, where the same authtoken will be seen
by both the staff client and Firefox browsing.
I willi also note that I rigged the wrapper to be easily adjusted for a
second, hard-coded host for selfcheck use, using oils://selfcheck instead of
oils://remote, that could be very easily turned into a minimal Firefox
extension. That would enable the selfcheck code to have permission to work
with printers and such without enablePrivilege (see next section).
2 - enablePrivilege Deprecation
The enablePrivilege security model has been removed and no longer functions,
and the staff client depended heavily upon it.
The Remote XUL fix provided us with a way to fix the enablePrivilege issues,
however. Because the Remote XUL has the Local XUL security context it
doesn't need enablePrivilege anymore. However, because enablePrivilege no
longer functions for those things *not* loaded through the wrapper some
things had to be changed.
For one, because if you are loaded through the wrapper you can't see
cookies, and if you can see cookies you can't do "XUL" things, the "XUL"
cookie has been removed. However, the wrapper can be detected by the
protocol you are using, so I added protocol checking to the isXUL checks I
know of. I am unsure about what, if anything, to do about the IAMXUL flag
that is set by the browser code, though.
For two, the JavaScript OPAC needs to be passed through the wrapper for some
staff functions to work currently. This has the additional side effect of
effectively forcing SSL mode for JSPac regardless of other settings.
For three, I went ahead and removed all enablePrivilege calls I could find
outside of the selfcheck code. As they no longer work they were just extra
cruft.
3 - Component changes
Component loading has changed significantly, and the easiest way to
accommodate the changes was to re-write our components. The command line
handler and force external components (the latter from another branch of
mine) were simple enough, especially as they aren't interacted with by other
code in the system (and as I wrote them I understood them anyway). While I
was at it I made the force external component more extension-friendly.
The data stash, however, was using a hack to accomplish what a singleton
component could do more easily. That and I couldn't get it to function as
it was. My solution was to re-write it as a singleton, removing the
interface definition that was not needed. As I changed how it functioned so
significantly I had to update all calls to it.
In addition, I hid a window loading routine in the data stash component for
easier loading of a *single* main Evergreen window from the Firefox
overlays.
4 - Extension changes
First off, Venkman, the DOM Inspector, and Chrome List are not guaranteed
compatible with XULRunner anymore. As such, I have removed the out of date
versions. Instead, I have added routines for loading Venkman and the DOM
Inspector as Extensions. This keeps them easy to update to later versions,
and makes us play much nicer when we are an extension. Also, now the
buttons/menu items for loading these interfaces disable themselves if the
extensions don't appear to be loaded.
A side effect of that was that Windows installers compressed with zlib were
unhappy. I switched to solid lzma, which in theory will get us smaller file
sizes too.
Secondly, Extensions have changed significantly in Firefox 4+, and Evergreen
had to be adjusted to deal with some of the changes. I adjusted the overlay
to take into account the new application menu, and I re-wrote the hotkey set
loading routines to deal with the fact that extensions don't get extracted
by default.
5 - Staff Client Build Process
XULRunner has some changes that needed accounting for, such as a root level
chrome.manifest being needed. Extension mode, as mentioned above, needed
tweaks as well, and other issues were run into along the way. Both XULRunner
and Firefox version support has been extended to 10.* and the base XULRunner
download has been bumped to 10.0.2.
In addition to the changes needed to support later XULRunner versions, I
also included alternate application name support, my fix for the
AccessControl plugin extraction, and support for OS-specific branding
folders. Mainly because I ended up using all of them in testing, and decided
to leave them in afterwards.
6 - Miscellaneous tweaks
In addition to everything else, while I was running through my various tests
I changed a few other things.
I enabled the navigation buttons on the OPAC interface embedded in the
patron overview for hold placement. With Remote XUL having the same security
context as Local XUL they now work.
I removed all non-SSL authtoken cookie assignments I could find, and flagged
all the SSL ones I could find as being "secure" cookies. On the subject of
cookies, I also told the logoff code to not erase every cookie, as those
running in extension mode might dislike all cookies going away. I know that
during my testing I disliked it, anyway.
I removed the code that messes with the user agent. Extensions are basically
not allowed to do that, and I don't think we need, or should rely on, a
modified user agent elsewhere, so adding a check specifically for when we
are not an extension didn't make a lot of sense to me.
I also made it so that the automatic login command line options will work
after a logoff (but they won't trigger a logoff) during a testing run where
I was logging off, clearing cache, and logging in again repeatedly.
To deal with CSS changes I changed most (but not all) popup elements to
menupop elements. Popups in later XULRunners are transparent by default.
In the reporter I found that "class" was being used as a variable name, but
that is a reserved keyword. I changed it to aClass. |
The initial commit message is reproduced below the link to the branch.
This is a significant enough change that I feel that more than just one person should review it. It has been being used on Jason Stephenson's development machine for a while now, though.
As a note, I included the work from the following bugs while building/testing the branch for this one:
Bug #921740 - External sites can be opened in the staff client
This has a component that needed to be re-written to support later XULRunner versions, and would *not* have worked in Extension mode in any satisfactory manner.
In addition, this covers Bug #628378 (Remote XUL support going away in Gecko 2) and may help correct Bug #491537 (xulrunner 1.9 (and Firefox 3.5) prints garbage to certain printers) by introducing a significantly later XULRunner version.
In the interest of documenting what has and has not been tested, I touched on the following:
Basic Circ Functions, Basic Admin functions (a number of config menus in particular), Automatic Updates, Installer Building, basic receipt printing, Vandelay loads (but I didn't test much beyond that), and minimal reporter testing. I touched on MARC View/Edit when errors were being generated there. Generally I bounced through random parts of the system I knew enough about to check things in, looking for anything that wasn't working, and tried to hit interfaces I didn't know enough about to ensure that they at least appeared to load without errors. I don't have a comprehensive list of them, though.
http://git.evergreen-ils.org/?p=working/Evergreen.git;a=shortlog;h=refs/heads/collab/tsbere/new_xulrunner
Commit message:
Improve Firefox/XULRunner Support
XULRunner/Firefox 4+ change the rules significantly, and to support them
significant changes have to be made in Evergreen.
This commit covers a number of things:
1 - Remote XUL Issues
Later versions of XULRunner disable support for Remote XUL without using a
whitelist entry, and whitelisting may stop working eventually as well. The
solution here is the addition of a new protocol wrapper that allows Remote
XUL to act like Local XUL.
The wrapper is oils:// and acts like HTTPS for the most part. The "host" the
client connects to is always "remote", however, and use of the wrapper has
multiple effects.
The first of these effects is the enabling of Remote XUL content, with as
close to the Local XUL security level as I could manage. This means that
nearly anything Local XUL can do, Remote XUL should be able to do too.
The second of these effects is the disabling of JavaScript-visible cookies.
XUL doesn't have cookies, and as such a side effect of the wrapper allowing
remote content to appear to be local content is the lack of cookies. This
does *not* affect server side visibility of cookies, however, so things like
TPac can still see cookies. This required fixing a lot of places that were
built to read data like authtokens out of cookies.
The third of these effects is the "hiding" of the hostname for anything
going through the wrapper. This prevents oils_persist from saving things
correctly. I thus made oils_persist check for the wrapper and use the data
stash to get the hostname instead. I suspect that we can get rid of the
location.hostname check entirely at this point, but have not done so.
The wrapper also sets a request header of OILS-Wrapper, value of 'true'.
That allows the server to know that the request came through the wrapper. I
used that ability to have TPac go into "Staff" mode when the wrapper is in
use. This will enhance extension use, where the same authtoken will be seen
by both the staff client and Firefox browsing.
I willi also note that I rigged the wrapper to be easily adjusted for a
second, hard-coded host for selfcheck use, using oils://selfcheck instead of
oils://remote, that could be very easily turned into a minimal Firefox
extension. That would enable the selfcheck code to have permission to work
with printers and such without enablePrivilege (see next section).
2 - enablePrivilege Deprecation
The enablePrivilege security model has been removed and no longer functions,
and the staff client depended heavily upon it.
The Remote XUL fix provided us with a way to fix the enablePrivilege issues,
however. Because the Remote XUL has the Local XUL security context it
doesn't need enablePrivilege anymore. However, because enablePrivilege no
longer functions for those things *not* loaded through the wrapper some
things had to be changed.
For one, because if you are loaded through the wrapper you can't see
cookies, and if you can see cookies you can't do "XUL" things, the "XUL"
cookie has been removed. However, the wrapper can be detected by the
protocol you are using, so I added protocol checking to the isXUL checks I
know of. I am unsure about what, if anything, to do about the IAMXUL flag
that is set by the browser code, though.
For two, the JavaScript OPAC needs to be passed through the wrapper for some
staff functions to work currently. This has the additional side effect of
effectively forcing SSL mode for JSPac regardless of other settings.
For three, I went ahead and removed all enablePrivilege calls I could find
outside of the selfcheck code. As they no longer work they were just extra
cruft.
3 - Component changes
Component loading has changed significantly, and the easiest way to
accommodate the changes was to re-write our components. The command line
handler and force external components (the latter from another branch of
mine) were simple enough, especially as they aren't interacted with by other
code in the system (and as I wrote them I understood them anyway). While I
was at it I made the force external component more extension-friendly.
The data stash, however, was using a hack to accomplish what a singleton
component could do more easily. That and I couldn't get it to function as
it was. My solution was to re-write it as a singleton, removing the
interface definition that was not needed. As I changed how it functioned so
significantly I had to update all calls to it.
In addition, I hid a window loading routine in the data stash component for
easier loading of a *single* main Evergreen window from the Firefox
overlays.
4 - Extension changes
First off, Venkman, the DOM Inspector, and Chrome List are not guaranteed
compatible with XULRunner anymore. As such, I have removed the out of date
versions. Instead, I have added routines for loading Venkman and the DOM
Inspector as Extensions. This keeps them easy to update to later versions,
and makes us play much nicer when we are an extension. Also, now the
buttons/menu items for loading these interfaces disable themselves if the
extensions don't appear to be loaded.
A side effect of that was that Windows installers compressed with zlib were
unhappy. I switched to solid lzma, which in theory will get us smaller file
sizes too.
Secondly, Extensions have changed significantly in Firefox 4+, and Evergreen
had to be adjusted to deal with some of the changes. I adjusted the overlay
to take into account the new application menu, and I re-wrote the hotkey set
loading routines to deal with the fact that extensions don't get extracted
by default.
5 - Staff Client Build Process
XULRunner has some changes that needed accounting for, such as a root level
chrome.manifest being needed. Extension mode, as mentioned above, needed
tweaks as well, and other issues were run into along the way. Both XULRunner
and Firefox version support has been extended to 12.* and the base XULRunner
download has been bumped to 12.0.
In addition to the changes needed to support later XULRunner versions, I
also included alternate application name support and support for OS-specific
branding folders. Mainly because I ended up using all of them in testing,
and decided to leave them in afterwards.
6 - Miscellaneous fixes
MARC View was building a data URL and dumping it into an attribute without
escaping it. Apparently this made XULRunner unhappy, so I added in escaping.
7 - Miscellaneous tweaks
In addition to everything else, while I was running through my various tests
I changed a few other things.
I enabled the navigation buttons on the OPAC interface embedded in the
patron overview for hold placement. With Remote XUL having the same security
context as Local XUL they now work.
I removed all non-SSL authtoken cookie assignments I could find, and flagged
all the SSL ones I could find as being "secure" cookies. On the subject of
cookies, I also told the logoff code to not erase every cookie, as those
running in extension mode might dislike all cookies going away. I know that
during my testing I disliked it, anyway.
I removed the code that messes with the user agent. Extensions are basically
not allowed to do that, and I don't think we need, or should rely on, a
modified user agent elsewhere, so adding a check specifically for when we
are not an extension didn't make a lot of sense to me.
I also made it so that the automatic login command line options will work
after a logoff (but they won't trigger a logoff) during a testing run where
I was logging off, clearing cache, and logging in again repeatedly.
To deal with CSS changes I changed most (but not all) popup elements to
menupop elements. Popups in later XULRunners are transparent by default.
Also for CSS changes, a menubar CSS entry in global.css was causing issues
with menu visibility on Linux. Things look fine without it, so I went ahead
and removed it outright.
In the reporter I found that "class" was being used as a variable name, but
that is a reserved keyword. I changed it to aClass.
To make things easier for url_prefix use I made it so that the urls hash is
checked by the url_prefix function. It looks for names up to the first of
several characters: /, ?, |. If urls contains that name it is replaced into
place. For | the | itself is removed in this process. |
|
