Re: unit testing C++ code from perl

I've glued a couple of Phlip's post together to make the reply all
in one chunk. If it got messed up in the process I apologise.

Phlip wrote:
Nick Keighley wrote:

do you have to read a book to understand any of this?

Unfortunately yes.

Beck's "Test Driven Development"? I was meaning to read it.



Are we supposed to just *know* what TEST or
TEST_FIXTURE does?

Nope. The authors of a *Unit should document such things. UnitTest++ has
1,300 downloads and no tutorial on their little home page, so these
probably represent people who already use CppUnit or CppUnitLite.

I'll take a look at CppUnit*


To write a general purpose library intended for a wide audience with[out] *any*
examples of usage or explanation of what the various parts do, just seems
bizzare to me!

Okay, there is the src/test folder. Read its Main.cpp, that's how you
launch the test run. Read any TEST() macro instance, that's how you write
a test case. So here's "Hello World":

TEST(whatever)
{
CHECK(false);
}

so macros arn't evil :-)


Noel and Charles might not feel motivated to spend their time writing a
manual that would be just the same as any other *Unit manual, but with
a few edits to match their own system. They are all generally the same.

a pointer to the other manual? Idon't know why I'm arguing with you.
it's
not your package!


Yes, I could puzzle it all out. But the temptation is to use the same time
to write my own.

And thus the problem. I wrote my itch-list because they formerly used my
rig, which had all these features that they neglected to understand or
use. So if you wrote your own, you could muddle through too, but you would
miss subtle techniques and conveniences that fully polished test rigs have.

but hacking code is such *fun*!


what! Its *good* to have no comments?! Not even a one liner saying what
it's *for*?

Ahem.

void ReportAssert(char const* description, char const* filename, int const
lineNumber)

Oookay. We have a function called ReportAssert, and it takes a
description. That's probably the complaint string for an error. "The
frob doesn't fit its Wibble socket", or whatever.

Then we have a filename and lineNumber. They are probably just the same as
the filenames and lineNumbers we have seen ever since we were little
kiddie-programmers.

Oh, and we didn't get creative with the names, like fileN or lNumber.
Those would have required comments. Like fileN // this is the filename,
and lNumber // this is the line number.

ok. I hate comments like that. I was arguing each *class* should have
1..few
lines saying what it's for. It's not always easy to tell what is for
use of
a user of a package and what is purely internal. I'm a hater of
unnecessary
and unhelpful comments, but *no* comments!!

I just looked at a file I created yesterday. It started with a one line
title
"a foo class for refrobnicating", Three lines explaining in more detail
what
it did and then an example of usage. I wasn't trying to flood with
comments
it just seemed natural.


Let's try another one.

template<typename ValueType>
void FormatToStream(MemoryOutStream& stream, char const* format, ValueType const& value)
{
char txt[32];
std::sprintf(txt, format, value);
stream << txt;
}

Hmm. That seems to format any primitive value as a string, then insert
it into a string. It probably wraps the "%2.0i" format flags that sprintf()
is famous for. And we are not production code, so we don't need
snprintf() or something more over-flow safe. But any comment I wrote
describing everything going on would be bigger than the function!

If you ask "what about the hard, complex methods?" that's the point -
there are none. All the functions are short, cheap, narrow, and obvious
like these. (There are those who claim unit testing helps generate such
code!)

Treat all comments as a failed opportunity to improve the code's clarity.

well we might have to agree to differ on this one...


BTW I wrote a tutorial for UnitTest++, here:

http://wiki.orbzone.org/TestFirstCiao

excellent, I'll give it a look.


Disregard the CIAO stuff. It would need ... lots of comments!


--
Nick Keighley

.

Relevant Pages

  • Help !! compile error with VS2005, but ok with VC6
    ... C3848: expression having type 'const CFormatter::ltwsz' would lose some ... @par Author: ... This class can be used for interpolating values into string, ... We might need to have an options to apply different format ...
    (microsoft.public.windowsce.embedded.vc)
  • Re: Timestamp to a string
    ... > concatenate a date string" which I think we can see it isn't. ... > of a date which doesn't mistake some legitimate filename for your ... I want to give an option of specifying the timestamp format to ...
    (comp.lang.java.programmer)
  • Re: Formula to convert yyyy.mm.dd to yyyy.mm.dd.ddd? (Perhaps a tricky one?)
    ... >>script set in old days to putting date at the required yyyy.mm.dd. ... >>But we've been needing the days in the filename, ... >Both results will be text string as Excel formatting cannot return the two day ... For the standard format one, ...
    (microsoft.public.excel)
  • Re: Error C2679
    ... came across a post saying that the <string> library wasn't included. ... CalculateBelowAverage(int const * const,float)" ... objects cin, cout, cerr, clog plus their wide counterparts. ... Read in the filename "C:\Documents and Settings\yourname\foo.barre". ...
    (microsoft.public.vc.language)
  • Re: DataSet XmlWrite error (c#) - URI formats not supported
    ... Based on error message, your string, which specifies location of the file to ... save is not correct or this format is not supported by DataSet. ... > System.Xml.XmlTextWriter..ctor(String filename, Encoding encoding) +34 ...
    (microsoft.public.data.ado)