Skip to content

Releases

PlexFixer 0.2.0 — off Electron

Choose a tag to compare

@focher focher released this 14 Nov 01:40

PlexFixer no longer runs on Electron. It renders in macOS's own webview and its
native side is Rust, which takes the download from 145 MB to 61 MB.

Why

The old build shipped 252 MB of Electron to run 2 MB of PlexFixer's own code —
three quarters of every download was the runtime. Nothing about the app needed
its own browser engine.

What did not change

Almost nothing moved to Rust. Rust does what only a native process can — spawn
ffmpeg, walk a filesystem, write the settings file, open a dialog — and
TypeScript keeps every decision. The analysis engine, the compatibility rules,
the H.264 SPS parser, the repair planner, the command builder and the job
runner are the same code, running in the webview, with the same tests.

Three things genuinely changed. The settings store, the library walk and the
environment probe are Rust now, with their own tests. The commands that start a
binary will only run the ones PlexFixer resolved — the bundled ffmpeg and
ffprobe, a path set in Settings, and your own ssh — so the page cannot ask
for anything else. And known_hosts is reached by name rather than by a path
from the caller, keeping the one file outside the app's own directory that it
writes to from becoming a general append.

Dolby Vision profile 5 can be converted

Profile 5 has no HDR10 base layer and keeps its picture in IPT-PQ-C2, so a file
relabelled 8.1 still shows the wrong colours — the familiar green/purple cast.
The only correct fix rebuilds the picture.

Settings → Compatibility → Dolby Vision profile 5 → Convert to HDR10 does
that: a full re-encode through ffmpeg's libplacebo, applying the Dolby Vision
RPU and writing plain BT.2020 PQ HDR10. The Dolby Vision layer is deliberately
dropped — that is the point, since an HDR10 file looks right on every client
where profile 5 only looks right on a Dolby Vision one. It is off by default,
forces 10-bit HEVC whatever your transcode codec says, and refuses outright
where libplacebo is unusable rather than quietly doing a plain re-encode.

The bundled macOS arm64 ffmpeg changed to one that has libplacebo, with Vulkan
and MoltenVK linked in. Intel Macs keep the old build and report the conversion
as unavailable rather than doing it wrong.

A default folder for the remote host

The remote browser opens where your library is instead of in the account's home
folder. Set it in Settings → Remote host → Default folder, or from the
browser itself with Start here next time — standing in the folder beats
typing a remote path from memory.

Smaller things

  • Settings is a list of sections with one on screen at a time, rather than one
    long scroll whose height changed under you when the remote host was toggled.
  • The library's tiles, search box and filter chips stay on screen; only the
    file list scrolls, and the table's column header sticks to the top of it.
  • The About panel reports the Tauri and webview versions in place of Electron,
    Chromium and Node.

Windows

This release is macOS Apple Silicon only. Tauri compiles a native binary
and does not cross-compile, so the Wine arrangement that produced a Windows
installer from a Mac no longer applies; that build now needs a Windows machine
or CI runner. An Intel Mac build needs an Intel Mac or a cross-build target set
up for it.

Opening it on macOS

PlexFixer is signed with a Developer ID certificate but is not notarized, so
macOS will refuse to open it on first launch. The DMG ships Allow PlexFixer
to open.command
next to the app: drag PlexFixer to Applications, then run
that script — it clears the download flag, checks the signature is intact, and
opens the app. By hand, the equivalent is:

xattr -dr com.apple.quarantine /Applications/PlexFixer.app

or System Settings → Privacy & Security → Open Anyway. macOS quarantines
the .command too, so if a Mac blocks the script as well, use one of those.

PlexFixer 0.1.18

Choose a tag to compare

@focher focher released this 13 Nov 21:21

Settings is a list of sections, not one long scroll

Every section used to be stacked in a single scrolling column. Anything that
changed how tall that column was moved everything under it — and turning the
remote host on and off does exactly that, because it adds and removes fields.
The scroll position then pointed at nothing, and the panel could end up
scrolled past its own content with a sliver of it left on screen.

The sections are now listed down the left — Languages, Repair modes,
Compatibility, Output, Transcoding, Remote host, FFmpeg — and only the one you
pick is drawn on the right. There is no height left to shift, and the list also
answers "which heading is that setting under", which a single long column never
did. Where you were last is remembered between visits.

Reset all settings to defaults sits under a divider at the bottom of the
list, apart from the sections, since it is not one of them.

Remote host reads in the order you use it

The host field now comes before the switch that turns it on; the switch is
disabled until a host is set, so the old order asked you to flip something that
could not be flipped yet. Port, private key and the remote ffmpeg/ffprobe paths
appear once there is a host for them to apply to.

The FFmpeg section now says plainly that its binaries are the ones on this
machine, and names the host the work is actually running on.

The library's header stays on screen

The tiles, the search box and the filter chips are how the list is steered, and
scrolling them off the top to reach the four hundredth row meant scrolling back
up again to change anything. They stay put now, and only the file list scrolls.

The table's own column header sticks to the top of the rows as they scroll,
which it never did before.

Opening it on macOS

PlexFixer is signed with a Developer ID certificate but is not notarized, so
macOS will refuse to open it on first launch. The DMG ships Allow PlexFixer
to open.command
next to the app: drag PlexFixer to Applications, then run
that script — it clears the download flag, checks the signature is intact, and
opens the app. By hand, the equivalent is:

xattr -dr com.apple.quarantine /Applications/PlexFixer.app

or System Settings → Privacy & Security → Open Anyway. macOS quarantines
the .command too, so if a Mac blocks the script as well, use one of those.

PlexFixer 0.1.17

Choose a tag to compare

@focher focher released this 13 Nov 06:41

Settings could be lost when two changes overlapped

Every settings change rewrote settings.json through one shared temporary
file. Two changes close together — flipping the remote toggle while a keystroke
in the host field was still saving — raced each other: the first rename moved
the temporary file away, the second failed, and that change was silently never
saved. In the worst interleaving a half-written file was renamed into place,
and because a file that would not parse was treated the same as no file at all,
the next save started from an empty object and every preference was gone.

Writes now queue behind one another and each uses its own temporary file. A
file that cannot be parsed is moved aside as settings.json.corrupt-<time>
instead of being overwritten, so the app still starts but the only record of
what your preferences were survives.

Related: PlexFixer no longer re-probes the environment on every keystroke. Each
probe is an SSH round trip to the remote host plus local test encodes, and a
half-typed hostname can only fail slowly. Typed fields now wait until the
typing stops; the remote toggle still takes effect immediately.

Bulk actions

One Bulk menu — Rescan, Repair, Force lossless remux, Force transcode —
replaces the buttons that used to appear with a selection and shift the
toolbar. The count and Clear selection are always there, greyed out when
nothing is selected, and Refresh all stays put.

A batch repair that would re-encode anything asks first, and says what each
file actually gets: Force transcode only re-encodes the files that need it,
and the rest are remuxed losslessly.

Backups that clean up after themselves

Delete backups after — 1 day, 7 days (the new default), 30 days or Never.
Each backup records when it was made, because a moved file keeps its old
dates. Expiry runs at launch, hourly, and after each repair, and waits for a
drive that is not mounted rather than giving up on it.

Settings → Remove all backups finds every backup folder under the library,
including ones from earlier versions, and confirms with the count and total
size before removing anything.

Header-only edits now take an instant APFS clone first where the volume is
APFS, so even the cheapest repair has something to go back to.

Leftover temporary files

A scan no longer lists a repair's in-progress .plexfixer-… file as a library
file. One that no running repair owns and that has not changed in 10 minutes
is removed, and the Library tab says it did.

H.264 levels: fewer needless re-encodes

ffprobe leaves the reference-frame count empty for most files, so the level
check assumed the worst case and sent streams labelled above 4.1 off for a
re-encode even when they used few enough reference frames to be relabelled
losslessly. PlexFixer now parses the SPS out of the stream's extradata and
reads the real max_num_ref_frames, so those files get the lossless relabel
they were entitled to.

When a stream genuinely cannot be relabelled, the finding now says which limit
it breaks — and notes when the MP4 header level (what Subler shows) disagrees
with the level inside the stream itself.

Smaller things

  • npm run dev now runs under a bundle named PlexFixer, so the macOS app menu,
    the Dock and Cmd-Tab stop saying "Electron".
  • The README documents the macOS signing situation plainly: these builds are
    signed but not notarized, and will not be. See below.

Opening it on macOS

PlexFixer is signed with a Developer ID certificate but is not notarized, so
macOS will refuse to open it on first launch. The DMG ships Allow PlexFixer
to open.command
next to the app: drag PlexFixer to Applications, then run
that script — it clears the download flag, checks the signature is intact, and
opens the app. By hand, the equivalent is:

xattr -dr com.apple.quarantine /Applications/PlexFixer.app

or System Settings → Privacy & Security → Open Anyway. macOS quarantines
the .command too, so if a Mac blocks the script as well, use one of those.

PlexFixer 0.1.16

Choose a tag to compare

@focher focher released this 13 Nov 01:45

Codec filters and a smarter idea of "blocked", plus a way past macOS Gatekeeper.

(0.1.15 was tagged but never published; everything in it is here.)

From Allen's PR #1

Library table. A Technical column showing size over the video codec and the default audio track's codec — HEVC, TrueHD +2 — with every audio track on hover, sortable by size.

Codec filters. Video and Audio pickers on the filter bar: dropdowns of checkboxes with counts. Codecs within one picker widen the selection, the two pickers together narrow it, and Audio matches any track — so DTS finds a file whose DTS track is not the default one.

Selection that holds. A Clear selection control beside the count, and a selected file that gets re-scanned or repaired into something the filters hide no longer silently drops out of the selection. Turning a filter on unselects nothing.

Blocked means blocked. Only a blocker, or a warning whose remedy is manual, now marks a file as blocked. A warning that only a re-encode would clear — "HEVC may not play on every client", a height over the ceiling — is advisory: the file counts as passed or fixable and the note stays in its details. The status, the Blocked chip, the Why column and the blocked-reasons panel all follow that one rule.

Repairs. An H.264 stream labelled with a higher level than its content needs is relabelled losslessly instead of re-encoded. Language codes are now written in the form each container expects — ISO 639-2/T in MP4, /B in Matroska — because writing /B into an MP4 made Apple players show French tracks as unknown.

Getting past Gatekeeper

Signing says who built the app; notarization is Apple having scanned it, and only that stops macOS refusing to open it. The build now runs scripts/notarize.mjs after signing, which submits each macOS app to Apple and waits for the ticket to be stapled. It is a no-op without credentials, and it fails the build if credentials are set and notarization does not succeed.

This build is not notarized, so the DMG carries Allow PlexFixer to open: drag PlexFixer to Applications, then double-click it. It clears the quarantine flag macOS attaches to downloads, checks the signature still verifies before touching anything, and opens the app. By hand that is xattr -dr com.apple.quarantine /Applications/PlexFixer.app, or System Settings → Privacy & Security → Open Anyway. Control-click → Open no longer works; Apple removed that in Sequoia. macOS quarantines the .command too, so it is a convenience rather than a cure.


macOS builds are signed with a Developer ID certificate. The Windows installer is unsigned.

PlexFixer 0.1.14

Choose a tag to compare

@focher focher released this 12 Nov 23:51

See where a repair has got to — and on which machine.

The problem with a percentage

ffmpeg's percentage covers one step of several. On a local disk the others go by too fast to notice; on the machine that holds the files they do not. Connecting, re-probing a finished file over the link, moving it into place across volumes — each can take minutes, and a bar sitting at 100% with no explanation is indistinguishable from a repair that has died.

What the queue shows now

Each job draws its steps as a track, with the current one marked and the percentage on it:

Connecting ✓ — Copying 55% — Verifying — Moving

The index move and subtitle sidecars appear as steps of their own when a file needs them. A job running on another machine is labelled with it (on plex@nas), and the library row's badge says the same in its tooltip.

Silence is measurable

Every job update now carries a timestamp, so a row that has reported nothing for 20 seconds says no update for 40s instead of leaving a stopped bar to speak for itself. That is the difference between a slow link and a dead one.

Honest about failure

A repair that fails is left standing at the step it failed in rather than marked finished, so nothing reads as verified or moved when it was not. A skipped job shows no track at all, because it never ran.

And ssh exit 255 is now recognised as ssh's own failure rather than ffmpeg's: a remote repair that comes back with it says the connection to the host dropped, and that nothing was promoted and the original is untouched — instead of reporting a mystifying encode failure.


macOS builds are signed with a Developer ID certificate but not notarized — right-click → Open on first launch. The Windows installer is unsigned.

PlexFixer 0.1.13

Choose a tag to compare

@focher focher released this 12 Nov 22:01

Run the work on the machine that holds the files — and get logged in to it.

(0.1.12 was tagged but never published; everything in it is here.)

Why run it there

A remux copies every byte of the file. Across a mounted share that is two network trips per byte — in to your Mac and straight back out to the server — which is most of the reason a repair that should take seconds takes minutes. Settings → Remote host moves the work to where the disk is: ffprobe and ffmpeg run there, and only the probe output and the progress come back.

Logging in

PlexFixer shells out to your own ssh, so your keys, your ~/.ssh/config, your agent and your known_hosts all apply, and the app never sees a credential. It never asks for a password — a prompt on a pipe would simply hang — so key-based login has to work on its own.

  • Private key: name one key, with a file picker that opens in ~/.ssh. Naming one sets IdentitiesOnly, so a host that refuses it says so at once instead of after ssh has offered every other identity it has.
  • First connection: the question ssh normally asks at a prompt gets asked in the connection test instead. On a host nobody has connected to before, the test fetches the key, prints every fingerprint with its type, and writes them to ~/.ssh/known_hosts only when you press Trust this host — then tests again in the same step.
  • A changed key is never offered as something to accept. If there is already an entry and the key no longer matches, the test stops and gives you the ssh-keygen -R line to run once you know whether that machine was rebuilt or something else is answering at that address. Keys are compared by their material, so a hashed known_hosts entry still matches, and trusting an already-trusted host is a no-op that simply re-tests.

The connection test

One scripted round trip reports what answered: the platform, ffmpeg and ffprobe with their versions and paths, and every POSIX utility the repair flow shells out to — ls, find, stat, df, mkdir, mv, rm, tail, head, base64, pkill. A missing one is named now rather than discovered half way through a repair on a machine you are not watching.

Browsing and scanning

With remote mode on, Scan… opens a browser for the remote filesystem — the native dialog can only see this machine — showing folders, media files and their sizes, with a path box for jumping to a known library root. A dropped local file now says it is on the wrong machine instead of failing file by file.

Everything downstream is unchanged: the same analysis, the same plans, the same write-to-temporary, re-probe, back-up and promote sequence, all happening on the far end. Encoder detection runs its test encodes there too, so a force transcode uses the server's NVENC rather than this laptop's idea of it.

Limits, stated plainly

  • The remote host must be POSIX — Linux, macOS or BSD, not Windows.
  • The in-place header edit (a few bytes rewritten instead of the whole file) stays local; a remote repair always rewrites the file.
  • Cancelling a remote repair stops the remote ffmpeg by matching the job's own temporary path, since closing an SSH connection does not reliably take the process with it.

macOS builds are signed with a Developer ID certificate but not notarized — right-click → Open on first launch. The Windows installer is unsigned.

PlexFixer 0.1.11

Choose a tag to compare

@focher focher released this 12 Nov 20:19

Measurements, a library you can sort, and blocked files that explain themselves.

Why was that remux slow?

Every repair now times its own stages — the check for an in-place edit, ffmpeg itself, the second pass that moves an MP4 index to the front, re-probing the output, moving the file into place, subtitle sidecars — and records which volume the source and the output are on, whether that is the same volume, and whether either is a network mount.

The Queue tab draws that as a bar per stage with the throughput and a realtime multiple, and says what the numbers point at rather than leaving you to interpret them: a copy running at the speed of a network share, an index move that is half the run, or a destination on another volume that forces the finished file to be copied again instead of renamed.

Copy diagnostics and Save diagnostics… collect the build, the ffmpeg in use, the settings that shape repair cost, the library breakdown and every measured repair into one plain-text block to paste into a bug report.

About

The app menu — and a button at the right of the title bar — opens a panel with the version, the platform, the Electron and Chromium under it, which ffmpeg is in use and where that binary came from, the hardware encoders detection actually found, and the settings folder. Copy details puts the same lines on the clipboard. The version now also appears in the per-file and per-repair reports.

A library you can sort and split

  • Every column header sorts: status, file, why, plan, size, scanned. The choice is remembered.
  • A Size column, so "which of these is the 60 GB one?" is one click.
  • Fixable and Blocked chips split the library by whether PlexFixer can repair a file at all. "Needs you" is now called Blocked throughout.

Blocked files say why

Opening a blocked row now answers a different question from "what is wrong": why nothing automatic will fix it, and what you can actually do. DRM cannot be remuxed by any tool. Dolby Vision profile 5 needs its RPU metadata rewritten, which ffmpeg alone cannot do and a plain re-encode would destroy. A file above the height ceiling would have to be scaled down, which throws away detail permanently — and Force transcode will not do it either, because the plan for such a file is still a stream copy. Each reason comes with the move that is open to you.

The Why column also now leads with the blocking finding rather than with a more severe one the repair already handles.

Queue

Jobs is now Queue, and Clear all is now Clear queue: it drops the repairs that have not started, while anything already running carries on and finished jobs stay as the record of what happened. Stopping a running repair is what Cancel is for.

Look

Deeper palette, lifted status tiles with a coloured edge, pill tabs, a sticky table header, tighter badges and a progress bar that moves.


macOS builds are signed with a Developer ID certificate but not notarized — right-click → Open on first launch. The Windows installer is unsigned.

PlexFixer 0.1.10

Choose a tag to compare

@focher focher released this 12 Nov 01:45

The repaired file now shows up

Refreshing a row after a repair updated the timestamp and nothing else. The
status still said Fixable, the reason still named the same fault, and the
job badge beside them said done — which reads as a repair that quietly
failed.

The refresh was right. If output placement is set to write beside the
original (name.plexfixed.mp4) or into an output folder, a repair never
touches the source file, so re-probing it finds exactly what it found before.
The repaired file is a different file, and nothing had ever put it in the
library: the fix that had just run was nowhere on screen, and only a re-scan of
the whole folder would turn it up.

A refresh now also pulls in the output of a finished repair when that output is
a file of its own. You get both rows — the source still reporting its faults,
which is the truth about that file, and the repaired one beside it saying it
direct plays. The completion message names where it went: written to
name.plexfixed.mp4
, on the job badge and in the Jobs tab.

Nothing changes for the default in-place repair, where the output is the
source and was already being refreshed, or for a container change, where the
new .mp4 replaces the old .mkv row rather than joining it.


macOS: the builds are signed but not notarized — right-click → Open the
first time.

FFmpeg: the bundled binaries are GPL builds, shipped as separate executables
and invoked as subprocesses, never linked. See NOTICE.md.

PlexFixer 0.1.9

Choose a tag to compare

@focher focher released this 12 Nov 01:00

Clear list

The Jobs queue could be emptied; the scanned list could not. Once a folder had
been scanned, its files stayed in the Library for good.

Clear list sits beside Scan… and forgets every scanned file. It asks
first, and the question says plainly what it does not do: nothing on disk is
touched — only PlexFixer's record of having looked at it. A fresh scan rebuilds
the list from the files themselves.

The search box, the status tile and the chips are cleared with it. They belong
to the rows they were narrowing, and leaving one on an empty library means the
next scan arrives into a list that is already hiding it.

A fix that came out of testing it

Clearing the list while Refresh selected was running put rows back — into
the table and into the stored library. Three of seven returned. A single-file
re-scan does not count as scanning, so Clear list stays live during one, and
a refresh of several files starts the later ones as the earlier ones finish, so
they outlived the clear and wrote their results into a list that no longer
existed.

A refresh now drops its result if the library no longer holds that path, which
is the right rule for every case and not only this one. An ordinary refresh is
unchanged.


macOS: the builds are signed but not notarized — right-click → Open the
first time.

FFmpeg: the bundled binaries are GPL builds, shipped as separate executables
and invoked as subprocesses, never linked. See NOTICE.md.

PlexFixer 0.1.8

Choose a tag to compare

@focher focher released this 12 Nov 00:02

Clear all

The Jobs tab has a Clear all button: it stops anything running and empties
the list. It asks first only when there is work under way — a queue of finished
jobs clears without a prompt. A job that is already running unwinds in the
background after the list is emptied, but it cannot reappear, and the file it
was working on is left exactly as it was, temporary file and all cleaned up.

Why a "simple" metadata change can take minutes

A language tag or a default flag is a fixed-width field in the MP4 header, and
PlexFixer edits those in place: 678 MB file, two bytes changed, 0.08 seconds,
nothing copied.

But if that file's index — the moov atom — sits at the end, the same repair
also has to move it to the front, and moving it rewrites every byte: the same
file, the same two bytes, 2.2 seconds on a local disk, and far longer across
a network share, where it scales with the size of the file and the speed of the
link.

The reason was being worked out and then thrown away, so the only visible
difference was the wait. The job log now names it, in plain terms, for every
repair that falls back to a full copy. It is a one-time cost per file: once the
index has been moved to the front, later metadata fixes on that file take the
fast path.

This also fixes a bug that was swallowing those explanations. The runner built
its "running" update from a snapshot of the job taken before the in-place
attempt, so anything logged while deciding was overwritten — including the
existing warning when an in-place edit fails and falls back to a full remux.
That warning has never actually reached the log in a shipped build.

Copy report, on any file

A report could only be copied for a finished job, which is no help with the
question that is actually hard: why did this file pass? Copy report now
sits beside Reveal in a row's details, for any scanned file, and emits the
whole picture — every stream ffprobe returned with its codec, tags, language and
dispositions, the findings or the absence of them, the planned actions, which
ffmpeg and ffprobe produced it, and the file's repair mode.

It also carries the settings that decided the verdict, because a report saying a
file is clean means little without them: "HEVC is compatible" and "MP4 is the
target" are settings, not facts about the file.

Fixes

  • Hide passed did nothing while the Passed tile was on, and looked no
    different for it — the checkbox was disabled, but the switch is drawn in CSS
    that never reflected that. It now always acts, and turns the tile off, since
    asking to hide passed files is the later instruction.
  • Choosing a new folder to scan clears the search box, the status tile and
    the chips. They belong to the view you were just looking at, and leaving them
    on means the files you asked for arrive into a list that is hiding them, which
    reads as the scan having done nothing. Refresh all keeps them, being a
    refresh of the current view rather than a new one.

macOS: the builds are signed but not notarized — right-click → Open the
first time.

FFmpeg: the bundled binaries are GPL builds, shipped as separate executables
and invoked as subprocesses, never linked. See NOTICE.md.