moved to it)
+ -go[no]dwarfsets switch to [disable]/enable generation of dwarf set
information, as this breaks gdb < 6.5 (used to be disabled in the
compiler by an ifdef, still off by default)
git-svn-id: trunk@8414 -
record variables is now independent of their packrecords setting
(except for packrecords C, which already calculated a reasonable
alignment). This means that e.g. a packed record consisting of two
pointers will be aligned at sizeof(pointer) normally. The internal
alignment of the individual fields of packed records obviously did
not change, also not if those fields are records themselves.
* The size of records without any packing atributes is also padded to
become a multiple of this improved alignment calculation, which
means that the size of such records may change after this patch.
Always explicitly specify a packing for records which are used for
data storage/transmission if you want to have a consistent layout.
git-svn-id: trunk@8409 -
* only suggest = if the user tried to overload <>
* suggest ** if the user tried to overload ^ (mantis #9606)
* no suggestion otherwise
(instead of always suggesting '=')
git-svn-id: trunk@8402 -
The downside is that because it is context-insensitive, several
(correct) optimizations which were performed in the past no longer
are now (and while some new ones are done now, the downside is bigger
-- but at least the code should be correct in all cases now)
git-svn-id: trunk@8385 -
TOption.TargetDefines and call it before parsing the config file,
so extra defines like UNIX and WINDOWS are already defined while
parsing the config file (mantis #9450)
git-svn-id: trunk@8371 -
perform the transformation of x:=f(hiddencomplexresult, ..)
-> f(x, ...) (the compiler now performs some very
conservative escape analysis for such types)
git-svn-id: trunk@8361 -
constants. The default is currently 32 bits/single, which corresponds
to the old behaviour (constants which cannot be exactly represented
in the default/chosen precision will also still be automatically
upgraded to higher precision). Supported constructs:
* Command line switch -CF<x>
* Compiler directive {$MINFPCONSTPREC <x>}
whereby in both cases <x> can be default, 32 or 64. 80 is not supported
because there is no generic way to figure out whether the current target
actually supports 80 bit precision floating point calculations while
parsing the command line switches (pbestreal can still change in case of
win64 or -Cfsse2)
git-svn-id: trunk@8349 -
source value to convert_l3, instead of only those with a destination
type whose size is < source size (like Delphi) + test -> fixes
toperator6 along with a host of wrong tordconst typeconversions in
the compiler sources themselves (although most are harmless)
git-svn-id: trunk@8323 -
negative cases caused by wrong automatic type conversion from
longint to unsigned tconstexprint (+ test for such jump tables)
* fixed darwin/ppc64 jumptables in case of jmptablenorange
git-svn-id: trunk@8311 -
converted to a pchar at compile time, remove the cstring_to_pchar node
because it expects a stringconstn in pass_generate_code
git-svn-id: trunk@8275 -
(widechar<->char, widechar<>*string), based on patch from
Rimgaudas Laucius (mantis #7758)
* no longer perform compile-time widechar/string->char/ansi/
shortstring conversions if they would destroy information
(they can't cope with widechars with ord>=128). This means
that you can now properly constant widechars/widestrings
in source code with a {$codepage } set without risking that
the compiler will mangle everything afterwards
* support ESysEINVAL return code from iconv (happens if last
multibyte char is incomplete)
* fixed writing of widechars (were converted to char -> lost
information)
git-svn-id: trunk@8274 -
