Hacker News Reader

Comment by Joker_vD | Hacker News Reader

> A static library is nothing but a bundle of object files, wrapped in an ancient and never-quite-standardized archive format.

To this day I'm astonished that it's not just tar. Or pax. Or even cpio! Or literally any file format that has any other use.

View on HN

Wasn't there first 'ar' as a general purpose archiver and later 'tar', "tape archiver"?

Why do you find it surprising that the archive format from that time was used to archive a bunch of files?

I wasn't alive but I'm pretty sure ar wasn't only used for this purpose in unix.

Because e.g. a.out didn't survive and was replaced? Several times, even?

That is a good point. Although I can only guess, it intuitively makes sense to me why the executable format would more quickly evolve than the archive format which "just works" still.

> Why do you find it surprising that the archive format from that time was used to archive a bunch of files?

It's surprising because we still use it today, not because it was used at the time.

Why would we break backwards compatibility, when it works just fine?

This response answers too much--why did we invent and adopt tar for all other use cases, then? Why do we not still use ar for its original general archiving purpose in any other situation but this one? Obviously there is some reason! I'm curious what that reason is. You can make regular archives with ar today but we don't--"it still works" wasn't good enough reason for any other archive use case. What are the general archiving requirements that don't exist for object archives that meant ar is still fine for this one use case?

Because we care about metadata and robustness for data files, but don't for some automatically generated artifacts that are consumed by the next program anyways? What do you need your static archives to have, that ar doesn't provide?

tar/pax are kind of terrible formats. They are hard to implement correctly. I'm glad they are not used more often.

cpio is pretty reasonable though.

zip is actually pretty great and I've been growing increasingly fond of it over the years.

The thing is, there is always tar(1) even in the most basic of distributions. And everyone uses tar.gz's or .bz2's or whatever for distributing all kinds of things, so tar is pretty ubiquitous. But the moment you want to do some C development, or any binutils-related, nope, install and use ar(1) which is used for literally one single purpose and nothing else. Because reasons.

Im not sure how ar does it, but tar has no centralised directory. The only way to get file 100 is to walk trough the 99 files before. This kills random access speed.

Ar puts a file called "/" as the first file of the archive. Inside, there is a number N, then a list of N file offsets, and then a list of N null-terminated strings. It's a symbol table of sorts: each null-terminated string is a symbol name, and the corresponding file offset points at the archive header for the object file that contains the symbol. The filenames themselves are not recorded centrally since it's not really needed.

At the time, ar existed and tar didn't. Tar came later.

At the time, a.out existed and COFF and ELF didn't. The switch from STABS to DWARF has also never happened, right?

> tar/pax are kind of terrible formats. They are hard to implement correctly. I'm glad they are not used more often.

I'll grant you "kind of terrible", but what's hard to correctly implement about tar? It's just a bunch of files concatenated together with a tiny chunk of metadata stuck on the front of each.

Having never done it myself, I don't know, but I do know that the "microtar" library I picked up off GitHub is buggy when expanding GNU Tar archives but perfect when expanding its own archives. Correctly creating one valid archive is a lot easier than reliably extracting all valid archives. The code appeared competent, I assume tar just has a bunch of historical baggage that you can get wrong or fail to implement.

Reply on news.ycombinator.com