On Jan 30,  4:54pm, jonathan@DSG.Stanford.EDU (Jonathan Stone) wrote:
-- Subject: Reworking ELF to be "normal-looking"

| [please cc: me on any replies]
| 
| 
| Does anyone know why NetBSD's userland ELF definitions are so
| different from the ones third-party code (eg., emacs) expects?
| 
| We use enums for the values defined for ELF header fields, with like
| ELf_pt_* for program headers, Elf_sht_* for section headers, etc.
| 
| Emacs, binutils (and other code? expects `standard' SVR4 names, and
| has to rename them to get otherwise-portable code to compile, viz:
| 
| # define PT_LOAD	Elf_pt_load
| # define SHT_SYMTAB	Elf_sht_symtab
| # define SHT_DYNSYM	Elf_sht_dynsym
| # define SHT_NULL	Elf_sht_null
| # define SHT_NOBITS	Elf_sht_nobits
| # define SHT_REL	Elf_sht_rel
| # define SHT_RELA	Elf_sht_rela
| 
| # define SHN_UNDEF	Elf_eshn_undefined
| # define SHN_ABS	Elf_eshn_absolute
| # define SHN_COMMON	Elf_eshn_common
| 
| emacs uses thees, plus HDRR, pHDRR and ElfW.
| 
| Are these mandated by ELF "standards", or just conventional?  What're
| the pros and cons of providing these (and the rest of the equivalents
| in libbfd's elf/common.h) in, say, sys/exec_elf.h?


I do, since I wrote them. At the time I wrote them, there was no standard.
I think that we should provide a <sys/elf.h> that has the standard names,
and eventually get rid of <sys/exec_elf.h>. Or have <sys/elf.h> include
<sys/exec_elf.h> and do the magic with the right defines. I like to use
enums better than #defines, because it makes it easier to debug things.

christos