The long-term plan is to test the builtin documentation with the binary output and cache it.

adapt tests accordingly

Michel Le Bihan authored 4 years ago and matu3ba committed 4 years ago

Fixes first part of #973

adding template description for merge requests

templates for issues and merge_requests for specific user information and better analysis of tradeoffs of certain features

Tom Almeida authored 5 years ago and AdminXVII committed 4 years ago

Before this patch, `let` only shows string and array variables, whilst
all mapping types are not shown. In addition, string and array variables
are separated from each other by comments.

This patch implements a new function on `shell::Variables` that returns
all the variables in scope and then modifies `Shell::list_vars` to print
all the variables. The syntax returned is equivalent to the method of
declaring the variables themselves, however this still results in
mapping types being indistinguishable from each other by looking at the
output of `let` (both `HashMap` and `BTreeMap` look the same).

Tom Almeida authored 5 years ago and AdminXVII committed 4 years ago

Why is this needed?
Well, at the moment nothing happens when the user tries something along
the lines of `let some_string[3] = 'v'`, even though we don't allow it.
The variable itself doesn't change, and we don't print any errors.
This patch just makes the function that would assign return an error
instead of silently failing.

Why not implement modifying strings by index?
Sure, we could do that, but it's not quite as obvious as it seems. We
support utf8 strings, so modifying the `i`th byte obviously doesn't
work, and could potentially make the string invalid. Another option
would be to set the `i`th character to whatever is passed in. This is
fine in theory, but it would be an `O(N)` operation in the average
case, and we'd have to shuffle all the following chars either further up
or further down the string.
While this may be a desirable ability to eventually have, it will
require significantly more work and thought around strings in ion than
simply returning an error.

Tom Almeida authored 5 years ago and AdminXVII committed 4 years ago

Prior to this commit, the following type was considered invalid:
    `[[float]]`

Of course, sometimes it is helpful to have arrays within arrays, and as
such, this patch allows for nested arrays. To do this, the array types
in `Primitive` have all been converted to a `Primitive::Array` type,
that holds a `Box<>` to another `Primitive`. There is - of course - a
slight performance penalty to be expected with moving to using another
`Box` for arrays, however after having run the benchmarks, the
difference appears to be negligible.

When some binaries set the TTY to non-blocking mode, the shell would panic because of a nonblock error. Fix this by reseting the O_FLAGS while reading lines

Avoid recompiling the regexset for each value, it is wasteful.
Add start and end anchor to match the full value with the regex by default

TODO: Does the regex crate have an option for this?

dardyfella authored 5 years ago and Michael Aaron Murphy committed 5 years ago

Fixes the issue where:

```
let arg = todo
echo -e "\n\n$arg"
```

Returns:

```

ntodo
```

Instead of:

```

todo
```

---

Fixes issue where:

```
echo -e "one\\\\\ntwo\\\\\nthree"
```

Returns:

```
one\\ntwo\\nthree
```

Instead of:

```
one\
two\
three
```