Bugzilla's Future. Much of this is the present, now. |
Bugzilla's future is a constantly-changing thing, as various developers "scratch an itch" when it comes to functionality. Thus this section is very malleable, subject to change without notice, etc. You'll probably also notice the lack of formatting. I apologize that it's not quite as readable as the rest of the Guide.
Bugzilla Blue Sky
Customisability
One of the major stumbling blocks of Bugzilla has been that it is too
rigid and does not adapt itself well enough to the needs of an
organisation. This has led to organisations making changes to the
Bugzilla code that need to be redone each new version of Bugzilla.
Bugzilla should attempt to move away from this to a world where this
doesn't need to occur.
Most of the subsections in this section are currently explicit design
goals for the "Bugzilla 3" rewrite. This does not necessarily mean
that they will not occur before them in Bugzilla 2, but most are
significant undertakings.
Field Customisation
Many installations wish to customise the fields that appear on bug
reports. Current versions of Bugzilla offer limited
customisability. In particular, some fields can be turned off.
However, many administrators wish to add their own fields, and rename
or otherwise modify existing fields. An architecture that supports
this would be extraordinarily useful.
Indeed, many fields work similarly and could be abstracted into "field
types", so that an administrator need write little or no code to
support the new fields they desire.
Possible field types include text (eg status whiteboard), numbers,
dates (eg report time), accounts (eg reporter, qa, cc), inter-bug
relationships (dependencies, duplicates), option groups (platform, os,
severity, priority, target milestone, version) etc.
Ideally an administrator could configure their fields through a
Bugzilla interface that requires no code to be added. However, it is
highly unlikely this ideal will never be met, and in a similar way
that office applications have scripting languages, Bugzilla should
allow new field types to be written.
Similarly, a common desire is for resolutions to be added or removed.
Allocations
?
Option Groups
?
Relations
?
Database Integrity
Furthermore, it is desirable for administrators to be able to specify
rules that must or should apply between the fields on a bug report.
For example, you might wish to specify that a bug with status ASSIGNED
must have a target milestone field that that is not untargetted. Or
that a bug with a certain number of votes should get ASSIGNED. Or
that the QA contact must be different from the assignee.
"Must" relationships could be implemented by refusing to make changes
that violate the relationships, or alternatively, automatically
updating certain fields in order to satisfy the criteria. Which
occurs should be up to the administrator.
"Should" relationships could be implemented by a combination of
emitting warnings on the process bug page, the same on notification
mails, or emitting periodic whine mails about the situation. Again,
which occurs should be up to the administrator.
It should also be possible for whine mails to be emitted for "must"
relationships, as they might become violated through direct database
access, Bugzilla bugs, or because they were there before the
relationship was enforced.
As well as implementing intra-bug constraints, it would be useful to
create inter-bug constraints. For example, a bug that is dependent on
another bug should not have an earlier milestone or greater priority
than that bug.
Database Adaptability
Often an administrator desires that fields adapt to the values of
other fields. For example, the value of a field might determine the
possible values of another field or even whether it appears (whether
it is "applicable").
Limited adaptability is present in Bugzilla 2, and only on the
"Product" field:
* The possible values of the target milestone, version and component
fields depend on the product.
* UNCONFIRMED can be turned off for specific products.
* Voting can be configured differently or turned off for different
products, and there is a separate user vote limits for each
product.
It would be good if more adaptability was present, both in terms of
all fields relying on the product, as well as the ability to adapt
based on the value of all fields.
Example ???
General adaptability raises the issue of circular references between
fields causing problems. One possible solution to this is to place
the fields in a total ordering and require a field refer only to the
previous fields.
In Bugzilla 2, changing the product of a bug meant a second page would
appear that allowed you to choose a new milestone, component and
version, as those fields adapted themselves to the new product. This
page could be generalised to support all instances where:
* a field value must or might be changed because the possible values
have changed
* is going to drop off because it it is no longer applicable, and
this should be confirmed
* must be specified because it is suddenly applicable, and the
default value, if one exists, might not be acceptable
Database Independence
Currently Bugzilla only runs on the MySQL database. It would be
desirable for Bugzilla to run on other databases, because:
* Organisations may have existing database products they use and
would prefer to run a homogenous environment.
* Databases each have their own shortcomings, including MySQL. An
administrator might choose a database that would work better with
their Bugzilla.
This raises the possibility that we could use features that are only
present in some databases, by appropriately falling back. For
example, in the MySQL world, we live without:
* record-level locking, instead we use table-level locking
* referential and record constraints, instead we checking code
* subselects, instead we use multiple queries and redundant "caches"
Multiple Front Ends
Currently Bugzilla is manipulated via the Web, and notifies via
E-Mail. It would be desirable for Bugzilla to easily support various
front ends.
There is no reason that Bugzilla could not be controlled via a whole
range of front ends, including Web, E-Mail, IRC, ICQ, etc, and
similarly for how it notifies. It's also possible that we could
introduce a special Bugzilla client that uses its own protocol, for
maximum user productivity.
Indeed a request reply might be returned via a totally different
transport method than was use to submit the request.
Internationalisation
Bugzilla currently supports only English. All of the field names,
user instructions, etc are written in English. It would be desirable
to allow "language packs" so Bugzilla can be easily used in
non-English speaking locales.
To a degree field customisation supports this, because administrators
could specify their own fields names anyway. However, there will
always be some basic facilities not covered by this, and it is
desirable that the administrator's interface also is
internationalisable.
Better Searching
General Summary Reports
Sometimes, the normal querying page leaves a lot to be desired. There
are other facilities already in place or which people have asked for:
Most Doomed Reports - All Bugs or All Bugs In A Product, Categorised
On Assignee, Shows and Counts Number of Bugs For Each Assignee
Most Voted For Bugs - All Bugs, Categorised On Product, Shows Top Ten
Bugs Voters Most Want Fixed
Number of Open Bugs For An Assignee - Bug List, Categorised On
Developers, Counts Number of Bugs In Category
The important thing to realise is that people want categorised reports
on all sorts of things - a general summary report.
In a categorised report, you choose the subset of bugs you wish to
operate on (similar to how you would specify a query), and then
categorise them on one or more fields.
For each category you display the count of the number of things in
that category. You can optionally display the bugs themselves, or
leave them out, just showing the counts. And you can optionally limit
the number of things (bugs or subcategories) that display in each
category.
Such a mechanism would let you do all of the above and more.
Applications of this mechanism would only be recognised once it was
implemented.
Related Bugs
It would be nice to have a field where you could enter other bugs
related to the current bug. It would be handy for navigation and
possibly even finding duplicates.
Column Specification Support
Currently bug lists use the columns that you last used. This doesn't
work well for "prepackaged queries", where you followed a link. You
can probably add a column by specifying a sort column, but this is
difficult and suboptimal.
Furthermore, I find that when I want to add a column to a bug list,
it's usually a one off and I would prefer it to go away for the next
query. Hence, it would be nice to specify the columns that appear on
the bug list (and general summary report) pages. The default query
mechanism should be able to let you specify your default columns.
Advanced Querying Redesign
?
Keywords
People have a need to apply tags to bugs. In the beginning, people
placed designators in the summary and status whiteboard. However,
these fields were not designed for that, and so there were many flaws
with this system:
* They pollute the field with information that was never intended to
be present.
* Removing them with a bulk change is a difficult problem that has
too many pitfalls to implement.
* You can easily get the capitalisation wrong.
Then dependencies were introduced (when?), and people realised that
they could use them for "tracking bugs". Again, dependencies were not
designed for that, and so there were more flaws, albeit different
ones, including:
* They aren't really bugs, so it's difficult to distinguish issues
from bugs.
* They can pollute bugs counts, and you must somehow exclude them
from queries.
* There is a whole lot of useless information on them. They have an
assignee but there is nothing to fix, and that person can get
whined at by Bugzilla. They have target milestones which must be
manually maintained. And so on.
Finally, keywords were introduced (when?) for this purpose to remove
the need for these two systems. Unfortunately, the simple keywords
implementation was itself lacking in certain features provided by the
two previous systems, and has remained almost unchanged since its
inception. Furthermore, it could not be forseen that in large
installations, the sheer number of keywords could become unwieldly and
could lead to a movement back to the other systems.
The keywords system was the right idea, however, and it remains so.
Fixing the keywords system is one of the most important Bugzilla
issues.
Bringing Keywords Up To Par
For the most part, keywords are very good at what they do. It is easy
to add and remove them (unlike summary/whiteboard designators), we can
simply see what issues are present on a bug (unlike tracking bugs),
and we do not confuse bugs with issues (unlike tracking bugs).
However, there are still some "regressions" in the keyword system over
previous systems:
* Users wish to view the "dependency forest" of a keyword. While a
dependency tree is of one bug, a dependency forest is of a bug
list, and consists of a dependency tree for each member of the bug
list. Users can work around this with tracking bugs by creating a
tracking bug and viewing the dependency tree of that tracking bug.
* Users wish to specify the keywords that initially apply to a bug,
but instead they must edit the bug once it has already been
submitted. They can work around this with summary designators,
since they specify the summary at reporting time.
* Users wish to store or share a bug list that contains a keywords
column. Hence they wish to be able to specify what columns appear
in the bug list URL, as mentioned earlier. They can work around
this using summary designators, since almost all bug lists have a
summary column.
* Users wish to be able to view keywords on a bug list. However
often they are only interested in a small number of keywords.
Having a bug list with a keywords column means that all keywords
will appear on a bug list. This can take a substantial amount of
space where a bug has a lot of keywords, since the table columns
in Bugzilla adjust to the largest cell in that column. Hence
users wish to be able to specify which keywords should appear in
the bug list. In a very real sense, each keyword is a field unto
itself. Users can work around this by using summary designators,
since they keywords will share the space in the summary column.
* Users wish to know when bugs with a specific issue are resolved.
Hence they wish to be able to receive notifications on all the
bugs with a specific keyword. The introduction a generic watching
facility (also for things like watching all bugs in a component)
would achieve this. Users can work around this by using tracking
bugs, as dependencies have an existing way of detecting fixes to
bug a bug was blocked by.
Dealing With The Keyword Overload
At the time of writing, the mozilla.org installation has approximately
100 keywords, and many more would be in use if the keywords system
didn't have the problems it does.
Such a large number of keywords introduces logistical problems:
* It must be easy for someone to learn what a keyword means. If a
keyword is buried within a lot of other keywords, it can be
difficult to find.
* It must be easy to see what keywords are on a bug. If the number
of keywords is large, then this can be difficult.
These lead some people to feel that there are "too many keywords".
These problems are not without solutions however. It is harder to
find a list of designators or tracking bugs than it is a list of
keywords.
The essential problem is it needs to be easy to find the keywords
we're interested in through the mass of keywords.
Keyword Applicability
As has been previously mentioned, it is desirable for fields to be
able to adapt to the values of other fields. This is certainly true
for keywords. Many keywords are simply not relevant because of the
bugs product, component, etc.
Hence, by introducing keyword applicability, and not displaying
keywords that are not relevant to the current bug, or clearly
separating them, we can make the keyword overload problem less
significant.
Currently when you click on "keywords" on a bug, you get a list of all
bugs. It would be desirable to introduce a list of keywords tailored
to a specific bug, that reports, in order:
* the keywords currently on the bug
* the keywords not currently on the bug, but applicable to the bug
* optionally, the keywords not applicable to the bug
This essentially orders the keywords into three groups, where each
group is more important than the previous, and therefore appears
closer to the top.
Keyword Grouping & Ordering
We could further enhance both the global and bug specific keyword list
by grouping keywords. We should always have a "flat" view of
keywords, but other ways of viewing the keywords would be useful too.
If keyword applicability was implemented, we could group keywords
based on their "applicability condition". Keywords that apply to all
bugs could be separated from keywords that apply to a specific
product, both on the global keyword list and the keyword list of a bug
that is in that product.
We could specify groups of our own. For example, many keywords are in
a mutually exclusive group, essentially like radio buttons in a user
interface. This creates a natural grouping, although other groupings
occur (which depends on your keywords).
It is possible that we could use collapsing/expanding operations on
"twisties" to only should the groups we are interested in.
And instead of grouping keywords, we could order them on some metric
of usefulness, such as:
* when the keyword was last added to a bug
* how many bugs the keyword is on
* how many open bugs the keyword is on
Opting Out Of Keywords
Not all people are going to care about all keywords. Therefore it
makes sense that you may wish to specify which keywords you are
interested in, either on the bug page, or on notifications.
Other keywords will therefore not bother users who are not interested
in them.
Keyword Security
Currently all keywords are available and editable to all people with
edit bugs access. This situation is clearly suboptimal.
Although relying on good behaviour for people to not do what they
shouldn't works reasonably well on the mozilla.org, it is better to
enforce that behaviour - it can be breached through malice, accident
or ignorance.
And in the situation where it is desirable for the presence or absence
of a keyword not to be revealed, organisations either need to be
content with the divulgence, or not use keywords at all.
In the situation where they choose to divulge, introducing the ability
to restrict who can see the keyword would also reduce keyword
overload.
Personal Keywords
Keywords join together a set of bugs which would otherwise be
unrelated in the bug system.
We allow users to store their own queries. However we don't allow
them to store their own keywords on a bug. This reduces the
usefulness of personal queries, since you cannot join a set of
unrelated bugs together in a way that you wish. Lists of bug numbers
can work, by they can only be used for small lists, and it is
impossible to share a list between multiple queries.
Personal keywords are necessary to replace personal tracking bugs, as
they would not pollute the keyword space. Indeed, on many
installations this could remove some keywords out of the global
keyword space.
In a similar vein and with similar effects, group keywords could be
introduced that are only available to members of a specific group.
Keyword Restrictions
Keywords are not islands unto themselves. Along with their potential
to be involved in the inter-field relationships mentioned earlier,
keywords can also be related to other keywords.
Essentially, there are two possibilities:
* a set of keywords are mutually exclusive
* the presence of a keyword implies another keyword must be present
Introduction of the ability to specify these restrictions would have
benefits.
If mutually exclusive keywords were present on a bug, their removal
would fix up the database, as well as reducing the number of keywords
on that bug.
In the situation where a keyword implies another keyword, there are
two possiblities as to how to handle the situation.
The first is automatically add the keyword. This would fix up the
database, but it would increase the number of keywords on a bug.
The second is to automatically remove the keyword, and alter queries
so they pick up the first keyword as well as the removed keyword.
This would fix up the database and reduce the number of keywords on a
bug, but it might confuse users who don't see the keyword.
Alternatively, the implied keywords could be listed separately.
Notifications
Every time a bug gets changed notifications get sent out to people
letting them know about what changes have been made. This is a
significant feature, and all sorts of questions can be raised, but
they mainly boil down to when they should be sent and what they should
look like.
Changes You're Interested In
As of version 2.12 users can specify what sort of changes they are
interested in receiving notifications for. However, this is still
limited. As yet there is no facility to specify which keywords you
care about, and whether you care about changes to fields such as the
QA contact changes.
Furthermore, often an unnecessary comment will go along with a change,
either because it is required, or the commenter is ignorant of how the
new system works. While explaining why you did something is useful,
merely commenting on what you did is not because that information is
already accessible view "Bug Activity".
Because of this unnecessary comment, a lot of changes that would
otherwise not generate notifications for certain people do so, because
few people are willing to turn off comments. One way to deal with
this problem is to allow people to specify that their comments are
purely explanatory, and that anyone who is not interested in the
change will not be interested in the comment.
Furthermore, one possible rationale for unnecessary comments is that
the bug activity does not display on the normal page and hence it is
difficult to cross reference comments and actions. Hence, it would be
beneficial to be able to do this.
Bugs You're Watching
Currently to receive a notification about a bug you need to have your
name on it. This is suboptimal because you need to know about a bug
before you can receive notifications on it. Often you are interested
in any bug with a field set to a specific value. For example, you
might be interested in all bugs with a specific product, component or
keyword.
If someone could automatically receive notifications about these bugs,
it would make everyone's lives easier. Currently the default assignee
and QA contact for a component will automatically receive
notifications for
Question: This moves half way to a BCC.
Bulk Changes
A very useful feature of Bugzilla is the ability to perform an action
on multiple bugs at once. However, this means that similar
notifications are currently generated for each bug modified.
This can result in a torrent of notifications that can annoy.
Furthermore, since the bugs are all changed close to each other in
time, it is easy for someone to mass delete all the notifications
generated by a bulk change and miss an unrelated notification in the
middle.
These factors can lead to a tendency for people to delay bulk changes,
or avoid them entirely. This is suboptimal.
It would be better if a bulk change generated only one notification
mail. This would vastly reduce the annoyance factor, and prevent
accidental deletion of notifications.
One problem with this change is that some people separate out
notifications using filtering. This means that they would no longer
be match parts of a bulk change under different filtering rules.
One possibility to resolve this is to allow people to specify groups
of bugs. All bugs within a group would go into the same
notification. The filters could then distinguish the different bug
groups.
In any case, it is likely there would need to be a transition period
to allow people to alter their filters.
Nominations
?
Linking Bugzilla Installations
The first example of linking Bugzilla installations together has is
the introduction of bug moving in version 2.12. However, it would be
useful to be able to link installations in more ways.
* Dependencies and other relationships between bugs in other
installations. This is difficult because dependencies are
synchronised on both bugs, so the installation that changes
dependencies would need to communicate the new state to the other
installation. It would also mean that relationships and
notifications that refer to other bugs would need to communicate
with the other installation.
* References to bugs in other installations. Currently if you type
"bug XXX" or "bug #XXX" where XXX is a number, you get an
automatic hyperlink to that bug. It would be useful if you could
say "YYY bug #XXX" where YYY is the name of another installation.
Retirement
?
Whiny Reports
?
Group Redesign
?
Hard Wrapping Comments
Currently Bugzilla "hard wraps" its comments to a specific line size,
similar to E-Mail. This has various problems:
* The way it currently works, wrapping is done in the browser at
submission time using a non-standard HTML extension not supported
by some (uncommon) browsers. These browsers generate comments
that scroll off the right side of the screen.
* Because comments are of fixed width, when you expand your browser
window, the comments do not expand to fit available space.
It would be much better to move to a world of soft wrapping, where the
browser wraps the text at display time, similar to a world processor.
And as in a word processor, soft wrapping does not preclude the
insertion of newlines.
Hard wrapping is too entrenched into text E-Mail to fix, but we can
fix Bugzilla without causing any problems. The old content will still
be wrapped too early, but at least new content will work.