Tcl Source Code

[677241c9b3] recently made changes to the internal utility routine
TclLog2() seeking optimization and confident extension to 64-bit range.

Conflicts encountered merging the changes to the dgp-refactor branch
revealed that branch holds considerably more improvements ready to
deploy.  A new branch dgp-msb has been opened to bring those changes over
to the trunk in a sequence of commits.  When the whole revision is complete,
I will be seeking review to approve accepting all the changes for merge to
trunk and backport to appropriate branches.

Accepted into 9.0 and 9.1 development.

Answering question about the non-support for an input of 0 to TclMSB().

Since the binary representation of 0 has no set bits, it is unclear
what the correct answer should be to the question what is the index
of the most significant set bit.

A return value of 0 is effectively an assertion that the input value 0
is no different from the input value 1.  Far from clear that's an
acceptable equivalence to all callers.

A return value of -1 succeeds in treating an input value 0 as something
unlike any other input.  However, most known callers of TclMSB() have the
intent of using the return value as an index, and a negative index is not
usually welcome.  Such callers would need to post-process the return value
to treat -1 differently.  This is no better than expecting the caller to
treat an input value of 0 differently before making the TclMSB() call. 
Either way, there's a testing branch needed if the caller knows a value
of 0 is a possibility.

FWIW, it is also true that this unsupported input value reflects the
same unsupported input value in the most efficient implementations.  The
limitations of clzll(), etc. are just showing through.

Have to take care at the input range boundaries to not
be testing undefined behaviors.  Thanks!  Should be better now.

All platforms except mine I guess.  Will look into it.

The testcases show this:

    ==== brodnik-1.0 TclMSB correctness FAILED
    ==== Contents of test case:

testmsb $v


---- Result was:
    -1
    ---- Result should have been (exact matching):
    0
    ==== brodnik-1.0 FAILED

Same failure on all platforms

I like the idea (and the implementation looks good)

I scheduled a build tonight, let's see what Github ACTIONS thinks.

Dopnald, this is clear wizard work. I even have no idea of the purpose. There are assembler intrinsics etc.

No way for me, sorry.

Take care, Harald

Anyone willing to look?

The dgp-msb branch is ready for an independent review of suitability to
merge to trunk.