tinydir_readfile returns -1 when file is a broken link

[asv22]

I.e. a file, which you could create with

ln -s /broken/link .

In that case iterate_sample.c shows directory listing only partially. And random_access_sample.c fails to open a directory at line 8.

tinydir_readfile finishes at line 559 when processing a broken link

My OS is Linux.

[lautis0503]

@asv22 Try the latest commit to see if it's fixed.

[asv22]

Thanks, for me it works great now.
It's worth noting that after the commit behavior of the library changed in this way:

before

If we have a link to an existing regular file, then file.is_dir=0, file.is_reg=1
If we have a link to an existing directory, then file.is_dir=1, file.is_reg=0

now

If we have a link to an existing regular file, then file.is_dir=0, file.is_reg=0
If we have a link to an existing directory, then file.is_dir=0, file.is_reg=0

[cxong]

interesting; I guess to find the properties of the link target, you need stat anyway.
I think we need to make a choice about how to support links

• if we pretend they are just real files then the existing behaviour is correct and we should update the samples to not break but continue when encountering broken links.
• If we show links we'll need to add a is_link attribute, and either
  • provide a readlink wrapper to show the path of the target, or
  • call stat again to see if the target is a file/folder, possibly recursively since links can target other links.

[lautis0503]

The second choice i will make.
Treat link files as something special and add a is_link attribute as in POSIX.1-2001.
is_dir and is_file should be 0.
We can provide new wrapper for read link files.

[wojdyr]

We had a related problem on Mac. The ifdefs before lstat enable it on Linux, but apparently not on macOS (which has lstat). We solved it by adding there

|| (defined __APPLE__)

[lautis0503]

I searched for manuals, and found lstat() is available in 4.2BSD.
Add the following for OS X and *BSD:

|| (defined__APPLE__) && (defined__MACH__)
|| (defined BSD)

[lautis0503]

Sorry for late pull request #70. I'll merge it unless it doesn't work fine for you.

[ccbenny]

Problems with symbolic links bit me recently in Tcl, which bundles Tinydir and Minizip, see https://core.tcl-lang.org/tcl/tktview/22ab2ae64ab2dd813719a985a6f125631ab1ba74 . I'd like this:

Treat link files as something special and add a is_link attribute as in POSIX.1-2001.
is_dir and is_file should be 0.
We can provide new wrapper for read link files.

Is that still the plan? It could be used in Minizip.

[lautis0503]

To be honest, we (at least myself) had no such thing called plan on this project. We will appear when any bugs or issues are found, because:

We have no ability to kill the bug, but the ability to kill the ones who raise the bug.

Just kidding.

I, myself, had never expected the project being used in any serious programs. (@cxong seems to use it in game, and I use it in personal project.) This is why when I made a dirty fix for this issue three years ago, although the API change is recognized, yet no actions were made to either fix the issue perfectly or prompt users to be aware of this API change. (If we used Semantic Versioning, the major version number should be increased.) Although that was a dirty fix, simply reverting the commit is another dirty fix, which make the original bug appear again.

More to say, in my opinion, if you want to use some program seriously, complete test suites or units should be used to make sure it all works. And this project is surely lack of that.

Thanks for your reporting, I would recently try to make this issue solved perfectly.

[cxong]

I think an easy partial fix is to add the is_link property, and leave it to users what to do with it
I'm not sure what happens if a link points to another link, and if you can have loops in links, and how to handle that elegantly.

[lautis0503]

We may also add a readlink procedure, which read the destination that link file points to, with parameters similar to readfile.
With is_link and readlink, users could easily write down code like while is_link, readlink again.
One thing to consider, MSYS/MSYS2/Cygwin/WSL are Linux (or POSIX?) layers upon the Windows system, would they process link file differently?

[lautis0503]

According to man pages of ln(1) of UNIX v7 and 4BSD

A hard link to a file is indistinguishable from the original directory entry; any changes to a file are effective independent of the name used to reference the file.

This means the hard links don't matter. We need to process the symbolic links.

According to man page of ln(1p) of POSIX.1-2008:

1. Switchs can be used to specify whether the source symbolic links are dereferenced when creating hard links.
2. Creating symbolic links doesn't care about whether the source files are symbolic links.

According to The GNU C Library Reference Manual, for version 2.36:

The macro MAXSYMLINKS specifies how many symlinks some function will follow before returning ELOOP.