|
From: Julian S. <js...@ac...> - 2011-10-21 04:15:33
|
On Thursday, October 20, 2011, Florian Krohm wrote: > On 10/20/2011 04:45 PM, Christian Borntraeger wrote: > >> +# Suppress warnings from Dwarf reader > >> +sed '/warning: evaluate_Dwarf3_Expr: unhandled DW_OP_/d' > > > > Hmmm, isnt this actually a bug, if gcc generated dwarf Ops that valgrind > > doesnt handle? > > Yes, it is. However, Julian looked at it and concluded that this would > be difficult to fix given the current dwarf infrastructure. > Suppressing these warnings is clearly less than ideal, but helps > filtering the regression that matter out of the noise. IIRC it was failing on DW_OP_piece (0x93 I think). This is related to variables that are partly in memory and partly in registers, eg a 64-bit value that has been partially spilled on a 32-bit CPU. The current infrastructure cannot represent them. For sure, hiding error messages with other DW_OP_ values isn't good, since we'll need to support them at some point (as Christian implies). J |