Hyperion Entertainment Message Boards

Dunno if i should post it to SDK sub-forum, or to general one (so i choice general one, but if it will fits better in SDK subforum , plz move it).

So, we have found back in time some small problem related to handling of errors of elf.library.kmod.

When elf binary have some unsupported reloc, aos spawn a window, with words "reloc #xxx is unsupported". And that XX value are not the same as in the elf specification. For example, when we have UNKNOWN_RELOC_36 (when binary compiled with enabled TLS), and by ELF specification, reloc 36 are R_PPC_SECTOFF - its not valid on aos4, because its reloc_72 by elf specification what mean R_PPC_TPREL16.

In other words, when developer see a window with error indication, he of course can't know that 36 its not 36, but 72.

That was founded with help of Frank Walle , who have old version of elf.library, in which: R_PPC_SECTOFF_HA is the last relocation which is known by name to elf.library (although unsupported). All the following relocations are called "Unknown_n" with n starting at 1! Yes, that means Unknown_1 is relocation #37 and Unknown_36 would be relocation #72 (R_PPC_TPREL16).

Will be cool if all the names will be included to os4 elf handler, and they will report the correct numbers of errors (and if some of them not supported, just skip them, but make final error report valid to elf specification).

Not big deal of course, but can drive to big troubles some developers who did't know that.

Hmm, weird... I thought I already did that, I know there is an array with reloc names in the library. Thanks for bringing it up, I'll check it.

@Hans-Joerg Frieden

Thanks for worring. Maybe it just fixed already in current beta-version, but not in the public update2 ?

Btw, as an idea: maybe will be not bad to add a bit more info for error, like for example Unsupported Reloc #XX (name_of_reloc_here). For now its just saying Unsupported Reloc #XX, and developer need to found name of reloc in the elf specification, but if add name of relococation itself (not only number) to the error message itself, then it will be twice better

kas1e wrote:@Hans-Joerg Frieden

Thanks for worring. Maybe it just fixed already in current beta-version, but not in the public update2 ?

Btw, as an idea: maybe will be not bad to add a bit more info for error, like for example Unsupported Reloc #XX (name_of_reloc_here). For now its just saying Unsupported Reloc #XX, and developer need to found name of reloc in the elf specification, but if add name of relococation itself (not only number) to the error message itself, then it will be twice better

Well, the "Unsupported Reloc" error is a weird thing to happen, because even a malfunctioning program should be able to be relocated. If you have one of these, then there is something very wrong and I doubt that any more information will help; mostly, it's a corrupt binary.

Well, the "Unsupported Reloc" error is a weird thing to happen, because even a malfunctioning program should be able to be relocated. If you have one of these, then there is something very wrong and I doubt that any more information will help; mostly, it's a corrupt binary.

Yeah, sure. I only meet with it 1 time and it was reloc36 (in reality 72), which was in the binary because i compiled code with enabled TLS. I also know that other developer (Jaokim, who works on jamiga2), also meet with it one time (with the same reloc36 (72) , and because of the same enabled TLS). Sure, its not very offten to meet with those errors, but it happens. Full and understanable error handling of any possible case are always good, because everything can happens