do not overwrite cuInfo->low, it's the base of a range

[oltolm]

The code overwrites cuInfo->low with a low from a range

            if( !hasLow || low<cuInfo->low )
            {
              cuInfo->low = low;
              hasLow = 1;
            }

and then uses cuInfo->low as the base of a range

            inline_info ii = { ptr,cuInfo->low,
              files,fileCount,callbackFunc,callbackFuncW,callbackContext,
              ptrOrig,(int)srcfileno+cuInfo->fileno_offs,lineno,columnno,
              cuInfo->fileno_offs };
            walkChildren( dbg,die,(ChildWalker*)findInlined,&ii );

That is not correct. The base of a range, according to spec, must be lowpc from a CU. This code is an optimization that has no influence on the runtime, I tested it. The patch simply removes the code.

I would also remove code like

      int hasLow = res==DW_DLV_OK && cuInfo->low;
      if( !hasLow ) cuInfo->low = 0;
      cuInfo->high = 0;

dwarf_lowhighpc does not change its out parameters in case of an error, so it's unnecessary defensive programming.

[ssbssa]

The would revert parts of 742f4ec, which I vaguely remember was needed for some debug infos I encountered.
And cuInfo->low/high have to be the minimum/maximum respectively, because they are later used to check if an address is contained in the CU.

[oltolm]

But the code right now overrides the lowpc from the CU. If the CU lowpc is 0, it will be overwritten with a non-zero low from a range. This is a common case. I could probably provide a test that reproduces the range base bug.

Why do you need the minimum and maximum when you can simply iterate over a range. Is it for performance reasons? I can provide them separately instead of overwriting the range base, although as I said I benchmarked it.

[ssbssa]

But the code right now overrides the lowpc from the CU. If the CU lowpc is 0, it will be overwritten with a non-zero low from a range. This is a common case. I could probably provide a test that reproduces the range base bug.

I wasn't aware that lowpc could be 0, so yes, I would like to see how to reproduce that.

[oltolm]

I pushed a new commit with the test to this branch oltolm@54c1842.

If you build with CMake, just run ctest, otherwise build test-cu-ranges.exe and a2l.exe and run test-cu-ranges .\a2l.