Wrong Content-Type for Adobe Illustrator and Microsoft Office files

[shunichi]

Some uploaded files have wrong content-type with carrierwave 3.0.3.

With carrierwave 3.0.3:

• Adobe Illustrator files (*.ai) are identified as application/pdf.
• Some Excel files (*.xlsx) are identified as application/zip. (Some Excel files are identified correctly)

With carrierwave 2.2.4:

• Adobe Illustrator files (*.ai) are identified as application/illustrator.
• Excel files (*.xlsx) are identified as application/vnd.openxmlformats-officedocument.spreadsheetml.sheet.

I know *.ai is a PDF with specific metadata and *.xlsx is a ZIP.
But *.ai should be downloaded as Adobe Illustrator file.

I think this commit cause the issue.
a2ca59c

It may be fixed if replcacing this line

carrierwave/lib/carrierwave/sanitized_file.rb

Line 324 in 8815592

Marcel::Magic.by_magic(io).try(:type)

with:

Marcel::MimeType.for io, name: original_filename, declared_type: declared_content_type

ActiveStorage identifies files in the same way.
https://github.com/rails/rails/blob/b5d63b9b16c6f251eaa08f61086f21ff3ef3292f/activestorage/app/models/active_storage/blob.rb#L353-L355

But I don't know it is a correct way to fix the issue.
It slightly changes the behavior in guessed_safe_content_type case. (in a case where Marcel can't detect the type from the file content and declared_content_type is nil)

[Juju-62q]

I have similar problem for with *.csv files.

With carrierwave 3.0.3:

• Comma-Separated Values files (*.csv) are identified as text/plain.

With carrierwave 2.2.4:

• Comma-Separated Values files (*.csv) are identified as text/csv.

With #2705 branch, *.csv files are now identified as text/csv.

[rajyan]

@mshibuya
Do you have some time looking into this issue?
This might be a blocking issue for some people upgrading to 3.x.

[mshibuya]

Closed by #2705.