Forums : Ohloh General Discussion

I had a look at the basic COCOMO definition, as suggested by the page Codebase cost.

Actually, I found the computation wrong.

My codebase estimation is 68 KLOC.
Currently, the raw line count (including licenses) of my project is 23KL.

This means that the codebase estimation is either wrong, or it is an historical summary.
Even if we consider the second option, which leads more or less to the appropriate COCOMO basic computation result, there is still a severe inconsistence with the COCOMO definition, which says:

KLOC per delivered lines of code.

while ohloh uses the historical ones.

Which is the problem?

Saverio

Saverio about 18 years ago

Hi Saverio,

I make no defense of COCOMO.

However, it sounds like the real problem is an incorrect KLOC. It looks like our new beta Subversion importer is making some errors. For instance, in revision 440, you deleted the entire /trunk directory, but our importer failed to subtract any lines of code. This is clearly an Ohloh bug. I will investigate.

Thanks,
Robin

Robin Luckey about 18 years ago

Hi Saverio,

We fixed a bug in our Subversion importer today, and I ran a new report for TinyHorrorSQL. This time we came up with 69,561 lines, which I believe is correct, and is the number you expected.

Let me know if you have any more questions,
Robin

Robin Luckey about 18 years ago

Hello,

I am sorry, but the count is still wrong.
The current line count, even counting a directory in 'tags' (about 1k lines), all the .txt/xml/html/noExtension, and the .java source files in 'trunk' (about 29k), it is still about 32klines, while the ohloh line count is 69k.

I had a look at the history graph (http://www.ohloh.net/projects/9806/analyses/latest) and seems like that the heavier factor is some kind of multiplication factor which grows the global line count to about 200-250%.
The proportion code/comments/blank seems like to be misaligned, in my estimation: couting 29K lines of raw .java sources, I can strip out 32(license) * about 130 files = 4000 lines, minus about 25% of the remaining, in comments and spaces, leads to about 20K lines of java code lines.

Actually, I had a look for example to SmallSQL (Java database) and its history is much more reasonable: about 21K code lines, which is more or less similar to a manual computation I've done in the past.
The important difference is that SmallSQL is CVS-based, while THSQL is SVN, so this could give some hint.

Bye!
Saverio

Saverio about 18 years ago

Hi Saverio,

I took a good look into this today and found the problem. It was another bug in our Subversion importer, which I have now fixed. I've scheduled a new report, which should be ready sometime tonight.

Not counting comments and blank lines, I'm now seeing about 18,000 lines of code (this includes both Java and HTML). I compared this with some manual counts and the results of sloccount, and I believe this is the correct number (finally!).

When the new report finishes, let me know if anything still seems amiss.

Thanks,
Robin

Robin Luckey about 18 years ago

You slashed my productivity to 30%! ;-)

Serisouly; the line counter is now accurate.
Wearing the QA hat ;-) I note that the complete count (in code statistics: http://www.ohloh.net/projects/9806/analyses/latest), including comments and blank linkes, is too high.

My current *.java line count of the trunk svn directory is currently about 23800 lines, while the statistics show about 31K. The HTML files amount to about 900 lines, which is negligible in respect to the difference.

I had a look at another project pseudo-randomly chosen (smallsql), and for that the total count is accurate (about 38K lines, accurately close to the 37.5K line count of its *.java files)

Is it possible that ohloh is (mistakenly, as supposed) counting also non-'trunk' (for example, 'tags') code? Including those, THSQL could reach about 30/31K.

Bye!
Saverio

Saverio about 18 years ago

Hi Saverio,

Yes, we are absolutely counting the tags directory, because it seems you requested it. If you want to count only the trunk, you need to change the enlistment URL to http://thsql.googlecode.com/svn/trunk.

It makes me cranky that you can't actually see the exact line counts by language on the web site (so much code to write, so little time), but here are the actual counts we are currently finding:

Java
Code 16940
Comments 9330
Blanks 4496
Total 30766

XML
Code 233
Comments 47
Blanks 53
Total 333

HTML
Code 828
Comments 0
Blanks 39
Total 867

Grand Total 31966

Unfortunately I can't separate out the number of lines that are coming from the /tags directory, although sloccount is telling me that the /tags directory contains 3557 lines of java code, so it probably contains a total of about 7K lines when you throw in the blanks and comments. Subtracting that would put us close to the 23800 you expect.

I appreciate your help here -- we've wrung some good bugs out of our importer so far, and I really want our line counter to be accurate.

Robin Luckey about 18 years ago

Thank you very much. This time was an overlook of mine. :-)
I changed the enlistement to point to /trunk and now the line count is accurate.

Glad to having been helpful!
Saverio

Saverio about 18 years ago