With the new HTTP.rb backend and the upcoming wget backend, there isn't
really a reason to ever ...

This was supposed to be part of the previous commit.

If the "time" standard library wasn't loaded by another piece of code,
the logging plugin would ...

Tempfile.new requires the basename to be provided on Ruby 2.2 or lower.

We don't need to assign the attacher class to a class variable on the
model, we can find the las...

Some people accidentally pass the :bucket option as nil in
Shrine::Storage::S3#initialize, for e...

ActiveRecord users that wish to translate Shrine validation errors along
with all other validati...

Closes #160

The download_endpoint doesn't use Down anywhere, so we don't have to
require it.

Base64.encode64 by default wraps the base64 result into 60 columns,
adding newlines after each 6...

In Mime::Types 3.0 the columnar storage became the default, so we can
stop requiring it explicit...

The MiniMime gem is significantly lighter than the Mime::Types in terms
of require time, so this...

See https://bugs.ruby-lang.org/issues/13539

Down 3.0 allows #size to be set on Down::ChunkedIO after it has been
initialized, and since Down...

This would be an unnecessary limitation, there is no reason why not to
give ability to instantia...

Unlike Base64.encode64, which adds a newline after each 60 characters
and a newline at the end o...

If we declare the File object as a static attribute value, the same File
object will be used for...

It turns out that `Base64.encode64(Digest#digest)` doesn't generate the
same output as `Digest#b...

Currently we are streaming the response body in chunks, and we're using
a buffer string so that,...

We do this for the same reason for which we made `Storage::S3#object`
public.

Fixes #142

This way users can convert the Rack uploaded file hash into an IO object
directly, which is usef...

With the upcoming `Shrine.rack_file` method users will be able to
convert Rack uploaded file has...

It's good practice for IO objects to conform to this interface.

It's useful for a user building their file upload flow to be able to
reuse functionality of Shri...

It's both more explicit and more readable, because it has the word
"attachment" in the name. Peo...

It's useful for developers building a file attachment flow to just be
able to just extract MIME ...

Generating hashes from files can be useful for doing checksums or
validating uniqueness of files.

It makes sense to me to return the resolved plugin module, especially
when testing in the console.

One user tried to return a hash of metadata with symbol keys in the
`add_metadata` block, and it...

This is in my opinion as secure as limiting on Nginx level, but it's
easier to maintain and more...

Passing regexes to extension/MIME type validations has been deprecated.

No need to pass the Tempfile path and have aws-sdk open its own file
descriptor, we can pass in ...

When FileSystem storage is used as the temporary storage, since its
Shrine::UploadedFile won't r...

By default aws-sdk will use chunk size of 5MB for multipart upload, and
50MB for multipart copy....

Also remove the Heroku note, as Heroku now seems to allow uploading to
any directory, not just t...

If instead of comparing by regular equality we switch to comparing by
hash equality, we can simp...

When migrating a large number of attachments, it might not be feasible
to make an HTTP request f...

Fixes #132

There is no need to load the `backgrounding` plugin only for a specific
uploader, and it complic...

When doing direct S3 uploads, correct content types should be set on
initial direct upload, beca...

Most image viewing desktop software automatically orient rotated images
(a photo can be rotated ...

Updates documentation wording

FileSystem#clear! with an :older_than parameter was entirely skipping
directories which were mod...

Using a Tempfile makes creating a DataFile significantly slower, because
2/3 of that time is spe...

RFC 2397 specifies that media type in a data URI can have additional
parameters, e.g.

data:te...

There is no reason to keep the parsed data URI in memory, so we save the
content to disk and dea...

This is allowed (or probably even required) by the RFC 2397 data URI
specification.

Performing a regex match on the whole data URI can be fairly slow, and
it is unnecessary conside...

So far we were both converting a data URI into an IO object and
assigning it to the model inside...

After we convert binary content to base64 format, we can deallocate the
binary string with Strin...

Grape converts all request parameters into a Hashie::Mash (which is a
Hash subclass), including ...