Am 06.10.2014 um 02:08 schrieb sebb:
On 5 October 2014 15:24, Felix Schumacher
<[email protected]> wrote:
Am 05.10.2014 um 14:35 schrieb sebb:
On 5 October 2014 13:26, Felix Schumacher
<[email protected]> wrote:
Am 05.10.2014 um 11:30 schrieb sebb:
On 4 October 2014 19:41, Philippe Mouawad <[email protected]>
wrote:
On Sat, Oct 4, 2014 at 2:10 PM, Felix Schumacher <
[email protected]> wrote:
Am 29.09.2014 um 22:32 schrieb Philippe Mouawad:
Hi Felix,
Hi
I agree with sebb, patch is interesting.
But it clearly needs to be documented (I think many users don't know
about
this feature which is really interesting) as long as code, reading
patch
first it wasn't clear for me what was intended.
I have added documentation to the patch and found two other things,
that
I
changed
in the same bug-entry.
The random order of applying the matchers, seems a bit strange, so I
sorted the matchers
first by their length and if the matchers are the same length, then by
the
name of their keys. So
the set
{'domain': 'example.com', 'server': 'www', 'regex': 'w.*' }
would be applied in the order ['domain', 'regex', 'server'] since
'domain'
has the longest matcher and
'regex' comes before 'server' alphabetically (matchers are both the
same
length).
Isn't it better to order by longest value or regexp ?
www is more specific than w.*
So would be :
domain, server , regex
Or the code could try to match every variable and select the one that
produces the longest match.
But rather than try and sort the regexes, which is always going to be
tricky to do "correctly" (whatever that means), maybe the user should
be given control of the matching order.
For example, it is probably possible to match by order of appearance.
It would certainly be possible to match the variables in sorted order by
name.
This would be a bit more awkard to use than changing the order of
variable definitions.
I just wanted to give a simple algorithm for ordering, which I think is
better than random ordering.
Correctness will be hard to implement, when everyone has a different view
on
the correct ordering.
I had thought of giving more control to the user by appending the
variable
names with something to sort by.
For example extending the above example with variable names ['domain',
'server', 'regex'] the names could be
changed to ['domain_3', 'server_1', 'regex_2'] to impose replacement in
the
order ['server', 'regex', 'domain'].
But what should we do with the suffix '_\d+'? (A prefix could be used,
too)
We could look for a specially named variable like '_regex_order' which
could
have a comma separated list of
the variable names in the wished order.
The longer I think about it, the more I am inclined to take the simple
ordering algorithm of length and then name. One can
always make any regex longer by adding useless junk like
'(?:WILLNOTBEFOUNDANYWAY)?' and in such a way influence
the order.
No, length of regex is not useful.
But it is easy to do and can be done consistently before trying to match :)
Just because it is easy does not make it useful.
More useful would be sorting by matched string.
I will try to do a patch which will do that, but I think it will be more
complex.
Yes, it will be more complex.
But I think it is more likely to be correct, but that's not
guaranteed, which is why I think the user should have control.
Sorting by name is awkward to use, and anyway what about non-regexes
that happen to match the same text?
Well in regex mode every string happens to be a regex. And with sorting by
name do you include using
(and possibly stripping off) a prefix or suffix?
It's not only regex matching that has potential ordering issues.
I don't think it's possible to automatically sort correctly by regex.
Well it is simple to order it correctly, when you want to have it sorted by
the current algorithm. But that is
obviously not your preferred order.
I just don't think it's possible to guarantee the correct order automatically.
The best one can hope for is that it will be right more often than
not, but there will always be edge cases.
Which is why I don't think it's worth trying to guess what people will need.
The user needs to be in control.
As I said, I think any repeatable
ordering is better then no order.
It needs to be predictable (and documented).
The current order is repeatable (on a given system), but is not easily
predictable.
So we should allow the user to control the search order, as I already
suggested a short while ago.
Right, what is your suggestion of means to accomplish that order?
I already suggested using the order of definition of the variables on
the test plan.
That should be possible, and is easier to use than variable renaming -
which may result in awkward names.
Having looked a bit deeper into this, it seems the order is already
defined :)
In Arguments#getArgumentsAsMap the arguments will be added to a
LinkedHashMap,
so we already have a deterministic behaviour.
Sorry for the noise on that one.
Regards
Felix
Would you like it to be another variable with a special name?
(I called that one '_regex_order' above).
Not unless it's not possible to use the definition order.
It's even more awkward to use than variable renaming.
What happens to variables, that the user missed to mention?
Hopefully not relevant, but I imagine they would be handled in alpha
order after the others.
Regards
Felix
If no one objects, I will submit it next week.
Regards
Felix
Thanks for contributing
Regards
On Monday, September 29, 2014, sebb <[email protected]> wrote:
On 29 September 2014 15:49, Felix Schumacher
<[email protected] <javascript:;>> wrote:
Am 29. September 2014 12:46:19 MESZ, schrieb sebb <[email protected]
<javascript:;>>:
On 29 September 2014 11:24, Felix Schumacher
<[email protected] <javascript:;>> wrote:
Am 29.09.2014 11:56, schrieb sebb:
On 28 September 2014 18:11, Felix Schumacher
<[email protected] <javascript:;>> wrote:
Am 22.09.2014 um 11:13 schrieb Marijn Wijbenga:
I've attached a jmeter project file and a html file that
demonstrates the
issue. In order to reproduce:
1. Load up xml-bug-test.jmx in jmeter.
2. Start the proxy (recorder)
3. Place xml-bug-test.html on a webserver somewhere (if on
localhost, do
not
forget to remove localhost from proxy exclusion if applicable)
4. Navigate with a browser to this file (using the proxy)
5. Click both buttons in order.
I could not post to a html file, hence the "test 2" button will
post to
Google. The page that loads has an error, but it still records
the
post
request which is what we want to see.
I also discovered that when I was using a "get" request instead
(I've
made
that "test 1") then it doesn't match the first character (%). I
think
this
is related.
The project has a user defined variable called "TEST" with a
value
os
".*",
I've ticked the box
To see the results, in the recording controller the last two
requests
contain a parameter with these values:
Test 1: %${TEST}
Test 2: <${TEST}>
Both should be just ${TEST} I believe.
In the current implementation the regex will be matched against
a
pattern
which looks like
\b(YOUR_VALUE)\b
As % and < are boundary characters they are excluded from you
pattern.
This is deliberate.
There were problems previously as partial values were being
unexpectedly matched.
See https://issues.apache.org/bugzilla/show_bug.cgi?id=52678
I thougt so. Maybe, that would have been helped by adding more
documentation, but then it is regex...
I would consider this a bug, or at least documentation could
be
a
bit
more
concise.
Patches welcome.
A patch was attached :)
I meant that we would welcome a patch for the documentation.
Or at least some indication of where the documentation needs to be
updated to clarify the current behaviour.
I will look into that.
Thanks.
What is your opinion on the option to detect parens and modify
the
regex
behavior?
Looks good to me.
The parens are very unlikely to have been used in existing tests, so
the modified behaviour is unlikely to break anything.
But we should document it in the release notes just in case.
Felix
Attached is a patch against trunk, which checks the regex if it
starts
with
'(' and ends with ')' and uses the regex as given, instead of
building
its
own version.
Please use Bugzilla for patches; it's easier to keep track of
them.
I have already done so yesterday shortly after sending my mail.
It
is
https://issues.apache.org/bugzilla/show_bug.cgi?id=57032
What is missing from the patch is documentation. If the feature
as
such is
ok, then I would add that to the existing documentation.
Regards
Felix
Also, see notes below.
-----Original Message-----
From: sebb [mailto:[email protected] <javascript:;>]
Sent: 21 September 2014 01:52
To: JMeter Users List
Subject: Re: Test Script Recorder XML Regex Matching
On 19 September 2014 16:45, Marijn Wijbenga
<[email protected] <javascript:;>> wrote:
Hi,
I have an issue, which might well be a potential bug, where a
posted
value
is
not being matched by the Test Script Recorder's Regex Matching
functionality.
The request I'm recording has a post value containing XML (SAML
token to
be
exact) which I'd like to replace with a variable automatically.
What does the value look like?
Does it have multiple lines?
No, it did not have multiple lines. I did check if this was the
case, but
it
wasn't
For testing purposes I have configured a User Defined Variable
(called
TEST)
with a value of "(?s)^.*$", I've tried "^.*$" and ".*" as well
(all
without
double
quotes).
Only ".*" replaces the content with this: <${TEST}>
That does not make sense.
".*" will match everything, including < and >, so the content
would
become
${TEST}
I know. It doesn't really. Hence I think this might be a bug.
I've tried other expressions as well and I'm able to match
anything
within
the
<> characters, but not those characters itself.
Again, that does not make sense.
The weird thing is, that inside the outer <> characters there
are
other
<>
characters that are matched fine. It's just the first and last
character.
Does anyone else have experienced the same thing, or is this a
known
issue?
It is not a known issue, and may not even be an issue.
Or should I post this in the developer's mailing list?
No, the developers all follow this list.
Great, please see attachment for an example.
Cheers
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
