Prefer MIME type when determining extensions for MediaBag items (#10557)

Currently, remote images added to the MediaBag are stored at paths with
extensions determined based on the external URI. For instance, an image
from https://example.com/image.png is stored as <hash>.png. If the
URI does not contain an extension (e.g., https://example.com/image),
then the content-type of the downloaded image is used to determine the
extension.

This change switches the precedence such that content-type is preferred
over extensions contained in the URI. This is necessary because some
images are located at URIs with misleading extensions -- shields.io,
for instance, serves SVGs from URIs with .yml extensions. With this
change, the image/svg+xml content-type is now preferred over the .yml
URI extension. This fixes a bug in the PDF writer in which such an image
would be mishandled due to not being identified as an SVG.

1 files changed, 5 insertions, 3 deletions

diff --git a/src/Text/Pandoc/MediaBag.hs b/src/Text/Pandoc/MediaBag.hs
index 678399703..1a4ea2ba1 100644
--- a/src/Text/Pandoc/MediaBag.hs
+++ b/src/Text/Pandoc/MediaBag.hs
@@ -107,9 +107,11 @@ insertMedia fp mbMime contents (MediaBag mediamap)
                   _     -> getMimeTypeDef fp''
   mt = fromMaybe fallback mbMime
   path = maybe fp'' (unEscapeString . uriPath) uri
-  ext = case takeExtension path of
-          '.':e | '%' `notElem` e -> '.':e
-          _ -> maybe "" (\x -> '.':T.unpack x) $ extensionFromMimeType mt
+  ext = case extensionFromMimeType mt of
+             Just e -> '.':T.unpack e
+             Nothing -> case takeExtension path of
+                             '.':e | '%' `notElem` e -> '.':e
+                             _ -> ""
 
 -- | Lookup a media item in a 'MediaBag', returning mime type and contents.
 lookupMedia :: FilePath