"SyntaxError: not a PNG file" in maps.map() call with 204 response

[dfloer]

I'm getting the following traceback (trimmed from pytest run) when pygbif gets a HTTP 204 error back from the GBIF API. According to the documentation, maps.map() returns a GbifMap object and doesn't raise any exceptions. What I'm seeing it it blowing up instead.

/pygbif/maps/map.py:172: in map
    return GbifMap(out)
/pygbif/maps/map.py:193: in __init__
    self.img = self.__prep_plot()
/pygbif/maps/map.py:203: in __prep_plot
    img = mpimg.imread(self.path)
/matplotlib/image.py:1501: in imread
    with img_open(fname) as image:
/PIL/ImageFile.py:121: in __init__
    self._open()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <PIL.PngImagePlugin.PngImageFile image mode= size=0x0 at 0x7FA7DA8A64F0>

    def _open(self):

        if not _accept(self.fp.read(8)):
>           raise SyntaxError("not a PNG file")
E           SyntaxError: not a PNG file
/PIL/PngImagePlugin.py:676: SyntaxError

Here's a command that triggers it for me now: pygbif.maps.map(z=10, x=10, y=10). It appears that the API is returning a 204 when there's no data to return for a query. Unfortunately, there's no upstream documentation on what HTTP status codes are allowed for the mapping API, so I puzzled out that this is probably when there aren't any occurrences of the taxon to show, but I also seemed to get it in possible error cases.

This is caused by not checking if there was actually a PNG received from the API before opening it:

pygbif/pygbif/maps/map.py

Line 192 in 1508ca9

if has(self.response.headers["Content-Type"], "png"):

The API responds saying that the content type is image/png (not png as the code above incorrectly assumes), but has no Content-Length header and a 0 byte body. The API returning an invalid PNG file is an API issue, it shouldn't do that, but that case should also be checked for.

Versions:
Python: 3.8.12 (old, needed for another dependency)
pygbif: 0.6.0
pillow: 8.3.2

[dfloer]

The monkey patch I ended up applying to work around #87 also removes the attempt to plot, so the allows me to properly get the "empty" response.