Documentation for 4.0
Guide to the TextTest Dynamic GUI
Administering interactive test runs
The dynamic GUI is selected on the command line via the “-g”
option. Alternatively, it is started by clicking “Run”
in the static GUI toolbar.
The initial left window is a tree view of all the tests that
are being run. These are colour-coded: white tests have not
started yet, yellow tests are currently in progress, green tests
are reckoned to have succeeded while red tests are reckoned to
have failed. By clicking the rows in this view details of what
happened to a particular test can be examined.
Ordinarily only failed tests will appear in this Test Tree view by default. You
can configure this via the config file and the status tab, see next paragraph.
Ordinarily, TextTest will run each test in order until all tests have
been completed, successfully or otherwise. You can however request that
it stops after the first test that fails, by selecting "Stop after first
failure" in the static GUI running tab, or "-stop" on the command line.
The initial right window is a summary of how many tests are
in a given state and is continually updated as they change. When
they complete, if they fail they are sorted into categories of
which files differed underneath the "Failed" category. Note
that as several files can differ in the same test this doesn't
necessarily add up to the total number of failed tests.
For each file which has differences, subgroups will then appear (labelled
Group 1, Group 2 etc.) for tests which have failed with an identical
difference in that file. Tests which have a unique diff will not appear
in any such group. "Identical" here is configurable: by default the "diff"
tool is used and TextTest will only compare using the lines beginning with '<'
or '>', filtering out the lines that indicate which line numbers were different.
So tests will be grouped together if they gain the same line at different points in the file,
for example. This can be configured using the config file dictionary setting
"text_diff_program_filters", which is keyed on diff programs and identifies lines from
their output that should be preserved in the output.
If a group is named e.g. "Group 2*3", that means TextTest is guaranteeing that the diff is the
same as the diff for group 2, repeated three times. Diffs that repeat are a common feature, for
example when a log statement is added to a program.
The status tab also doubles as a way of selecting and hiding whole
categories of tests at once. Clicking the lines will select all
such tests in the test tree view. Unchecking the check boxes on
the right will cause all tests in that category to be hidden in
the test tree view on the left. If you uncheck one of the
“<file> different” lines, note that tests will
only be hidden if all
files that were different are marked as hidden, so if a
file is different in more than one file it will not be hidden
until you uncheck all the files where it was different. In this
example there is only one file anyway.
Note that whether categories are visible by default can be
configured using the config file list entries "hide_test_category" and "show_test_category".
By default everything under "Failed" is shown, and everything else is hidden.
To refer to the various categories, you can use any name that appears in the status tab,
or you can use the keys from for
example the "test_colours" entry in the personal preferences table.
The grouping described above is mostly good enough when your system has only changed one thing at once. It's clearly good development
and testing practice to avoid changing many things before testing but sometimes it may be unavoidable, for whatever reason.
In this case finer-grained information can be useful, by allowing viewing, grouping and approving of parts of files. This is done with the
"Split Result Files" action in the test popup and Actions menu.
It will be greyed out unless you tell it how to split your files. You should find a suitable marker in your result files which delineate
different "sections" of the file. For example, when GUI testing with StoryText, each action in the GUI generates its own section in the
output file, so we might want to split on these sections. We do this via:
[file_split_pattern]
stdout:'(.*)' event created with arguments
This implies that when doing "Split Result Files" on the "stdout" file, each line that matches the given regexp will be taken as part
of a new "section". The 'group' indicated by the brackets will determine how the section files are named. You must provide such a group.
On doing "Split result files", the file in question will be split and the various "subfiles" appear in the File View. These will then
be compared as normal and grouped in the Status tab. This allows further identification of repeating patterns, even in files which may
contain other differences.
Approving any of the "subfiles" will result in the main file being reconstituted with the approved version of that section and the previously
stored version of all the other sections.
Note that the split is just a convenience for viewing and approving results, it does not affect how the results are stored and will
not affect any future runs of the tests.
When tests are selected via either of the above means, a new
“Test” tab is created on the right, containing the
details of what's going on in that test. Another tree view, this
time of the result files of that test, appears in the top right
window, while a textual summary of what happened for this test
can be seen in the “Text Info” tab below. The files
are also colour-coded, depending on whether TextTest thought
they were different (red) or not (green).
Double clicking on files from the test view will bring up
views of those files, using the external tools specified in the
config file. The relevant entries are "diff_program"
for viewing differences, "follow_program" for
following a running test and "view_program" for
viewing a static file. These default to “tkdiff”,
“xterm -e tail -f” and “emacs” respectively on
UNIX systems, “tkdiff”, “baretail” and
“notepad” on Windows. By default differences will be
shown if they are thought to exist (red files) otherwise the
file is simply viewed. To select other ways to view the files,
right-click and select a viewer from the popup menu.
"follow_program" is used for monitoring the progress of a running test. Ordinarily it
is only available from the popup menu, but you can make it come up when double clicking
a file in a running test by setting "follow_file_by_default" to 1. When starting this viewer,
TextTest will set the environment variable TEXTTEST_FOLLOW_FILE_TITLE to an appropriate name
for the window title, which can then be referred to in your "follow_program" setting, as indeed
the default setting on UNIX does.
Note the config entries described above are all “composite
dictionary” entries and can thus be configured per file
type, using just the stems as keys. It is thus easy to plugin
custom viewers if particular files produced by the system are
more conveniently viewed that way. For example, this will cause the "stdout"
files to be viewed with "xemacs" rather than the default "emacs":
[view_program]
stdout:xemacs
On the Text Info tab you will see a textual preview of all
the file differences, along with a summary of what happened to
the test. This textual preview is also used by the batch
mode email report. The tool used to generate the diff can be
configured via the config file entry “text_diff_program”
(it defaults to “diff”, which is also used
internally by “tkdiff”). Each file diff is
truncated: the number of lines included can be set via
“lines_of_text_difference”, which defaults to 30.
If several files differ they will be displayed in
alphabetical order. Sometimes this is not ideal as some files
give better information than others. In these cases you should
use the config file dictionary setting
“failure_display_priority” which will allow you to
ensure the most informative file comes at the top. Low numbers
imply display first.
To protect from very large files being compared (or viewed),
which may result in using all the machine memory,
you can specify a maximum file size for this, using the
“max_file_size” config file entry. It defaults to "-1" which means no limit.
For example:
[max_file_size]
diff:1GB
emacs:500MB
I.e. the keys are the various programs. "default" can be used as normal to set a fallback
limit for tools not explicitly listed. If a file exceeds
this size for "diff", no diff report will be displayed in the Text Info tab, and if you try and view such a file
from the GUI you will get a warning that the file is very large before it will be loaded.
When tests fail, you can examine the differences as above,
and sometimes you will decide that the change in behaviour is in
fact desirable. In this case, you should "Approve" the
test, or group of tests. This operation will overwrite the
permanent "approved" files with the "temporary"
files produced by this run. It was known as "Save" up until TextTest 3.26.
To achieve this, the dynamic GUI has an “Approve”
button in the toolbar and the File menu, and also an “Approve As...” action
in the File menu. Select the tests you wish to
approve from the left-hand test window by single-clicking, and
using Ctrl+left click to select further tests. (Press Ctrl-A to
select all tests) On pressing “Approve” (or Ctrl+S),
all selected tests will have their stored (approved) results overwritten.
On approving a test, by default all files registered as
different will be saved. You can however approve only some of the
files by selecting the files you wish to approve from the file view
(under the Test tab) in much the way you select multiple tests
in the tree view window.
TextTest will normally save the files exactly as the system
under test created them, and then perform its filtering on them again before each test run. You can however tell it to save the filtered output as the new approved result,
by setting the configuration setting "save_filtered_file_stems".
This is mainly useful when
- the information filtered away isn't interesting and
- the filtering is much simpler to do on the present run than on an arbitrary one
To enable this for e.g. "stdout" and "myfile", you set
save_filtered_file_stems:stdout
save_filtered_file_stems:myfile
By doing "Approve As...", you get a dialog up with
further configuration options. You can configure which version the results are stored as (look
here
for a description of versions).
By default, they
will overwrite whatever file they have compared with, irrespective of what version markers it has, and will not
have any version suffix if they are new files. This is signified by the default "<existing version>" option,
and is also the behaviour of "Approve". There is also the special option "<full version>" which will store the tests
with whatever version they were run with (this corresponds to the default behaviour with TextTest 3.20 and earlier).
Sometimes you don't want
results to be saved for particular versions, this can be
configured via the “unsaveable_version” entry which
will cause these versions not to appear in the list in future, or be used when approving with "<full version>".
You can also specify a version identifier to back up the old results as, for which there
is a text field in the "Approve As..." dialog. This is particularly useful when your project
has made a release and you have a separate version of the tests for that release.
You can also overwrite all the files, even if they were
regarded as the same, via the option “replace successfully
compared files also”: this is a way to re-generate the
run-dependent text
for all files in a failing test. Note that this only works on tests where at least some files were different, it won't work on tests that
are considered to have succeeded. To systematically regenerate run-dependent text, run with the "--ignorefilters" flag
(or check the box in the Advanced tab in the static GUI) which will cause tests to fail and hence make the process of approving
run-dependent text much more visible.
There are several occasions on which it can be useful to trigger a recomputation
of the test results. Sometimes other versions of the tests are simply approved, which changes files shared between several
versions. Sometimes the approved files for the tests are changed externally, via something like
a version control update. The filtering settings in the config files might also be changed, and hence
a recomputation could be triggered to see if the test now succeeds with the new filtering rules. For
these reasons there exists a "Recompute Status" action in the Actions menu.
Each time tests are selected or approved in the dynamic GUI, TextTest will
check the file modification times to see if a recomputation seems to be necessary. If it is, a small refresh
icon will be shown to the right of that test in the test tree view. It isn't necessary for such an icon
to be shown in order to perform "Recompute Status" but in most cases it won't have any effect.
It's also possible to run this action on a test that has not yet completed. In this
case it will perform a filtering on the test files produced so far, which gives you the possibility to view them
and check if the test is clearly going to fail. It will also attempt to estimate how far the test has progressed,
by comparing the size of the approved files to the size of the temporary files produced so far.
Sometimes you have a lot of tests failing for different reasons.
It can be helpful to be able to manually classify these tests as you discover what caused individual failures, so that
you can see which tests still need to be checked. You can therefore "mark" tests
with a particular text, which will cause them to be classified differently and be easy
to hide as a group from the Status view. This is achieved by right-clicking on the test
and selecting the relevant item from the popup menu.
On the right hand side of the dynamic GUI there is a tab called Run Info. It contains two textual displays.
The top half shows information about the TextTest run itself, what arguments were given and when it started etc.
The bottom half shows information about the run of the currently selected test(s), including the exact command line
and environment it was started with. The purpose is twofold: to aid with identifying different dynamic GUI windows
if many runs are started simultaneously on longer-running tests, and to help with debugging if TextTest appears not to be running
the system under test in the way you expect.
For especially important longer runs it's possible to provide a name for the run, which is a way
of providing a succinct description and a guard against accidental closure at the same time. This is done via
View / Set Run Name from the menus in the dynamic GUI. It can also be done externally via "-name" on the command line
or via the "Name this Run" field in the Running /Advanced tab in the static GUI.
If the run is named in this way, the window header will be changed, the Run Info updated and when you try
to quit a warning dialog will appear to make sure you haven't made a mistake.
There is an environment variable TEXTTEST_DYNAMIC_GUI_INTERPRETER which is a hook to provide additional arguments
to Python when starting the dynamic GUI from the static GUI. This is mostly useful for development, for
example to plug in StoryText in the self-tests. It's no longer necessary to use this to measure coverage, which
works via a Python interpreter startup hook now.
In TextTest 3.27 the icons were redesigned. Naturally there are always conservative people who don't like such change.
If you are one of them, you can get the pre-3.27 icons back by adding the setting "retro_icons:1" to your personal config file.
|