Name:           neuro-mime-types
Version:        1.0
Release:        6%{?dist}
Summary:        Shared MIME type definitions for neuroimaging and electrophysiology formats

License:        MIT
URL:            https://gitlab.com/morgan.hough/neurofedora
Source0:        neuro-mime-types.xml
Source1:        neuro-mime-types.LICENSE

BuildArch:      noarch

BuildRequires:  libxml2
BuildRequires:  shared-mime-info

# The file triggers in shared-mime-info rebuild /usr/share/mime on install
# and uninstall, so this package needs no scriptlets of its own -- but it is
# useless without the database, so the dependency is hard.
Requires:       shared-mime-info

%description
MIME type definitions for the file formats used in neuroimaging and
electrophysiology, so that desktop environments and file managers can identify
them and offer the applications that read them.

Covers volumes (NIfTI, CIFTI, MGH/MGZ, MINC, AFNI, MRtrix), tractography and
surfaces (TCK, TRK, TRX, GIFTI), 3D Slicer scenes (MRML, MRB), and
electrophysiology recordings (FIFF, EDF,
BioSemi BDF, GDF, BrainVision, XDF, NWB, Blackrock, Neuralynx, Open Ephys,
CTF MEG, Intan, Plexon, Spike2, Axon ABF), plus diffusion gradient tables.

Detection uses file content where a reliable magic number exists and the file
extension otherwise; each entry records which, and why, in a comment.

Formats already defined elsewhere are deliberately left alone: DICOM by
shared-mime-info, and NRRD, MetaImage and Radiance .hdr by f3d.

%prep
%autosetup -c -T
cp -p %{SOURCE0} %{SOURCE1} .

%build
# Nothing to compile. The validation that matters happens in %%check.

%install
install -Dpm0644 neuro-mime-types.xml \
    %{buildroot}%{_datadir}/mime/packages/neuro-mime-types.xml

%check
# 1. Well-formed XML. Necessary, nowhere near sufficient: the file that
#    shipped a root-XML element without its required namespaceURI attribute
#    was perfectly well-formed.
xmllint --noout neuro-mime-types.xml

# 2. The real gate: build a MIME database from it and fail on ANY diagnostic.
#    update-mime-database exits 0 while printing errors, and a rejected
#    element does not stop the others from loading -- so the missing
#    namespaceURI above silently dropped the <glob> that followed it, and
#    .gii files fell back to application/xml. Nothing about that looks like a
#    failure unless you are reading stderr, which is why this greps it.
mimedir=$(mktemp -d)
install -Dpm0644 neuro-mime-types.xml "$mimedir/packages/neuro-mime-types.xml"
update-mime-database "$mimedir" 2>"$mimedir/err.log" >/dev/null || :
#    The pattern matters: a missing attribute prints "Error in type ...", but a
#    malformed comment prints "... parser error : Double hyphen within comment"
#    and then "Failed to parse", at which point the ENTIRE file is dropped and
#    the first pattern never matches. Both spellings are checked because both
#    have now happened here.
if grep -qiE '^error|Error in type|parser error|Failed to parse' "$mimedir/err.log"; then
    echo "update-mime-database rejected part of the definition:" >&2
    cat "$mimedir/err.log" >&2
    exit 1
fi

# 3. Every declared type must reach the compiled database. A type that fails
#    to compile leaves no trace in the install tree, so count both sides.
declared=$(grep -c '<mime-type type=' neuro-mime-types.xml)
compiled=$(find "$mimedir" -name '*.xml' -path '*application*' | wc -l)
echo "declared=$declared compiled=$compiled"
test "$compiled" -eq "$declared"
rm -rf "$mimedir"

%files
%license neuro-mime-types.LICENSE
%{_datadir}/mime/packages/neuro-mime-types.xml

%changelog
* Mon Oct 05 2026 Morgan Hough <morgan.hough@gmail.com> - 1.0-6
- Define application/vnd.nrrd and application/vnd.mhd, which f3d also
  defines, deliberately and with identical globs. 3D Slicer's two desktop
  entries declare both types as handled; nothing here defined either, so on
  a machine without f3d the declarations were INERT and a .nrrd resolved to
  nothing -- Slicer was never offered for its own native format. Requested
  by the 3dslicer session, approved by Morgan, conditional on the test
  below. 33 types.
- Tested before adding, because 1.0-5 removed a duplicate definition for
  exactly the opposite reason. The rule is not "never define a type twice",
  it is "never define it twice DIFFERENTLY": update-mime-database merges
  definitions that agree. A database built from f3d's real
  f3d-image-formats.xml plus this one resolves, identically with and
  without f3d present and with rc=0:
      .nrrd .nhrd .nhdr -> application/vnd.nrrd
      .mhd  .mha        -> application/vnd.mhd
- *.nhdr is one glob beyond f3d's set and is the reason this is worth
  doing at all. f3d carries *.nhrd, a transposition; NRRD's detached
  header, which is what Slicer writes when it separates a header from its
  .raw, is .nhdr. Nothing in Fedora or in this COPR claimed it, and
  3dslicer ships four files that handle it (confirmed independently by the
  3dslicer session). Both spellings are listed so the f3d-installed case
  stays consistent. Upstream report drafted at
  docs/upstream/f3d-nrrd-nhdr-glob-typo.md for Morgan to file.
- ci/mime-agreed-duplicates.tsv records the two types as intentional
  duplicates so ci/check-mime-consistency.py does not report them, and the
  allowlist suppresses ONLY that report: a glob mapping to two different
  types is still a hard failure, which is what would catch the two
  definitions drifting apart later.
* Sun Oct 04 2026 Morgan Hough <morgan.hough@gmail.com> - 1.0-5
- Stop defining application/x-mrtrix and application/x-mrtrix-tracks. The
  mrtrix3 package defines both, so every install of the two together had one
  type defined twice, by two files, disagreeing: our *.mif.gz glob carried
  weight 60 and mrtrix3's own application/x-mrtrix-gz carried the default
  50, so .mif.gz resolved to application/x-mrtrix -- a type mrview.desktop
  does not declare. Installing both packages broke a file association that
  either one alone got right. 31 types.
- Considered and rejected: adding NRRD and MetaImage. f3d defines
  application/vnd.nrrd and application/vnd.mhd in f3d-image-formats.xml, so
  claiming them here would have recreated the same bug this release fixes.
  f3d was not installed on the machine the check first ran on, which is why
  ci/mime-external-owners.tsv now records externally-owned types explicitly
  instead of inferring them from whatever is installed locally.
* Fri Oct 02 2026 Morgan Hough <morgan.hough@gmail.com> - 1.0-4
- Revert the *.hdr claim added in -3, at Morgan's direction: the two-file
  form is not worth contending for, because its *.img half cannot be typed
  by anything. The extension belongs to everything, and shared-mime-info
  already gives it to application/vnd.efi.img.
- The ni1 magic stays, so a two-file header is still recognised when its
  extension is not already spoken for, but we no longer game the glob
  weights to take *.hdr off image/vnd.radiance.
- Recorded while here, because it is easy to get backwards: Analyze 7.5 is
  not supported and cannot be. Its .hdr carries four ZERO bytes where NIfTI
  puts its magic -- verified by writing one with nibabel -- so an Analyze
  header is indistinguishable from any other 348-byte file. Nothing in this
  package ever matched Analyze; ni1 is NIfTI-1's own two-file form, a
  different format wearing the same two extensions.

* Fri Oct 02 2026 Morgan Hough <morgan.hough@gmail.com> - 1.0-3
- Recognise the OLD two-file NIfTI form. .hdr/.img is not just Analyze 7.5:
  NIfTI-1 inherited it, and such a header carries "ni1\0" at offset 344
  where the single-file form carries "n+1\0". The magic for it was already
  here and never fired, because f3d claims *.hdr for image/vnd.radiance.
- The fix is not a higher magic priority. Measured at 60, 80 and 90, all
  three lost: when exactly ONE type claims an extension, that glob decides
  and the content is never consulted. Claiming *.hdr ourselves makes the
  glob match AMBIGUOUS, which is what sends the question to the magic --
  and weighting it 40, below Radiance's default 50, means a .hdr matching
  no magic still falls to Radiance. Verified both directions: a nibabel
  written pair header resolves to application/x-nifti and a real
  "#?RADIANCE" file still resolves to image/vnd.radiance.
- *.img stays unclaimed and cannot be otherwise: it is raw voxels with no
  header, so nothing in it identifies the format, and shared-mime-info
  already types the extension as application/vnd.efi.img. Typing the .hdr
  is enough in practice, since every reader opens the header and finds its
  own .img.
- tests/neuro-mime-types.bats covers all of this, with the Radiance control
  synthesised when f3d is absent so the collision is always exercised.

* Tue Sep 29 2026 Morgan Hough <morgan.hough@gmail.com> - 1.0-2
- Add application/x-mrml and application/x-mrb, the two formats 3D Slicer saves
  scenes in. Slicer is the largest application in this repo and had no desktop
  integration of any kind; these are the half of it that belongs here.
- The .mrml entry is a collision, not a new extension. shared-mime-info already
  defines text/x-mrml for the unrelated Multimedia Retrieval Markup Language
  and claims *.mrml, so both globs sit at weight 50 and the extension cannot
  decide. XML is case sensitive and the two formats differ in exactly that: a
  playlist opens "<mrml ", a Slicer scene opens "<MRML ". Magic at priority 70
  over a 0:512 offset range, since an XML declaration shifts the root element.
  Measured both ways: a saved scene resolves to application/x-mrml and a
  hand-written playlist still resolves to text/x-mrml.
- root-XML is declared too but does not settle it alone: the namespace table is
  consulted only once a file is known to be XML, which is after the glob tie.
- .mrb is a Zip container with nothing distinguishing in its header, so it is
  glob-only and sub-class-of application/zip, which keeps an archive manager
  reachable as a secondary action.
- Harden %%check: the parse-failure gate matched "Error in type" but not
  "parser error"/"Failed to parse". A double hyphen inside an XML comment hit
  exactly that path -- the whole file was rejected, no types compiled, and
  update-mime-database still exited 0. Caught here by the declared-vs-compiled
  count, which is the backstop the grep is meant to make unnecessary.

* Mon Sep 28 2026 Morgan Hough <morgan.hough@gmail.com> - 1.0-1
- Initial package. 34 MIME types for neuroimaging and electrophysiology.
- Written because the COPR had three MIME packages across 1046 binary
  packages (afni, f3d, paraview) and none of them covered the formats the
  field exchanges: a .nii.gz on Fedora 44 opens in GNOME Text Editor.
- Magic numbers are verified against real files where possible and marked as
  such in the XML: NIfTI "n+1\0" at offset 344 with sizeof_hdr 348, read from
  ds000114; the FIFF 12-byte tag header 00000064 0000001f 00000014, identical
  across all seven .fif files in ds000117. Entries with no verified magic say
  so and use the extension.
- Two collisions resolved on content rather than on extension: *.bdf is
  application/x-font-bdf in shared-mime-info, and BioSemi BDF takes it back
  with magic at priority 90; *.nev is claimed by both Blackrock and
  Neuralynx, and both globs are weighted below default so the magic decides.
- Formats left to their existing owners: DICOM (shared-mime-info), NRRD,
  MetaImage and Radiance .hdr (f3d). *.img and *.hdr are not claimed at all:
  they collide with disk images and Radiance respectively, and a viewer that
  wins that fight has taken something from the user.