Non existent path causes panic instead of Error result

[chupaty]

New versions of tera panic instead of returning Error when passed an invalid glob path. For example:
rsgen-avro

let mut tera = Tera::new("/dev/null/*")?;

Problem caused by use of unwrap() instead of ? operator, for example:

tera.rs
let parent_dir = std::fs::canonicalize(parent_dir).unwrap();

[Keats]

That's not good, I'll fix it today unless someone beats me to it

…

On Wed, 15 Mar 2023, 02:23 chupaty, ***@***.***> wrote: New versions of tera panic instead of returning Error when passed an invalid glob path. For example: rsgen-avro <https://github.com/lerouxrgd/rsgen-avro/blob/9e3eb0bfd0f7ba7a583bb7f3166fd67c41f4e708/src/templates.rs#L338> let mut tera = Tera::new("/dev/null/*")?; Problem caused by use of unwrap() instead of ? operator, for example: tera.rs <https://github.com/Keats/tera/blob/226d0108cdb64c8d056e46c5c9a67a4a4e8549ea/src/tera.rs#L135> let parent_dir = std::fs::canonicalize(parent_dir).unwrap(); — Reply to this email directly, view it on GitHub <#819>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFGDI2UPDC2CXBK4QZ25KLW4EKXVANCNFSM6AAAAAAV3FAIN4> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[vimpostor]

@Keats

Your fix caused a regression in behaviour (well to be fair, the behaviour regression was already present in #799).
Before this patch series, tera would allow invalid globs but expand them to the empty set. Now they always lead to an error.

This feature was useful, because in zola it was possible to remove the templates directory completely of a static website and just use the templates directory of the theme in use.

Now with this change, this is no longer possible and zola always requires the templates directory to exist (it can be empty though). This is because zola calls Tera::parse(&tpl_glob).context("Error parsing templates from the /templates directory"), which will now error with this change. Before the change it would just internally expand the invalid glob to the empty set, which is completely fine, because zola extends it later with the theme's template directory and with zola's builtin templates.

To me this change doesn't make any sense, because it is inconsistent:

• If /templates exists and is an empty directory, /templates/* will expand internally to the empty set and return Ok()
• If /templates doesn't exist, then /templates/* will return an error

In my opinion both cases should have the same behaviour, i.e. both cases should expand to the empty set, and not return an error just because an internal implementation detail (std::fs::canonicalize) requires the path to exist.

Please see the following patch for a simple fix, that will return to the old behaviour but still with the fix from #799 present. If you agree, I can also make a PR.

diff --git a/src/tera.rs b/src/tera.rs
index bf29f6d..1aece99 100644
--- a/src/tera.rs
+++ b/src/tera.rs
@@ -132,7 +132,10 @@ impl Tera {
         // See https://github.com/Keats/tera/issues/574 for the Tera discussion
         // and https://github.com/Gilnaa/globwalk/issues/28 for the upstream issue.
         let (parent_dir, glob_end) = glob.split_at(glob.find('*').unwrap());
-        let parent_dir = std::fs::canonicalize(parent_dir)?;
+        let parent_dir = match std::fs::canonicalize(parent_dir) {
+            Ok(d) => d,
+            Err(_) => std::path::PathBuf::from(parent_dir),
+        };
         let dir = parent_dir.join(glob_end).into_os_string().into_string().unwrap();
 
         // We are parsing all the templates on instantiation
@@ -1225,6 +1228,6 @@ mod tests {
     fn doesnt_panic_on_invalid_glob() {
         let tera = Tera::new("\\dev/null/*");
         println!("{:?}", tera);
-        assert!(tera.is_err());
+        assert!(tera.is_ok());
     }
 }

Otherwise if you insist this is the correct behaviour in tera, this can also be fixed from zola's side.

[Keats]

I think it makes sense and is consistent for me to return an error if a path doesn't exist? templates/* can perfectly be empty if there are no files, however if the folder doesn't exist I'd want to know rather than silently ignoring whatever is given.

[vimpostor]

Your opinion is perfectly reasonable, it just doesn't match what most other languages do. In Bash it can be controlled with the failglob option, which is off by default. Similarly Python's glob.glob("/dev/null/*") returns the empty list instead of failing. Also note that in Rust globwalk doesn't error as well, the error simply comes from the usage of std::fs::canonicalize.

Maybe it's better to fix the problem in zola.

[T3kla]

I have spent 2 hours searching why the hell my Zola project wasn't building to GitHub Pages. Turns git wasn't pushing the empty templates folder.