The section offered "an OpenPGP public key or an S/MIME certificate" and
only the first half had ever been tried. Every probe behind it used
OpenPGP keys, and every message the registry returns names OpenPGP --
including for input that is not OpenPGP at all -- so the server reads as
though OpenPGP were the only format it knows. Shipping the claim on that
evidence would have been a guess dressed as a feature, which is the one
thing this section is written not to do.
It holds. Confirmed live on 0.16.20 (2026-09-05) with a self-signed
X.509 certificate carrying emailProtection and an email: SAN:
registered, read back, destroyed. And Stalwart parses it as seriously as
it parses OpenPGP -- a malformed certificate is refused by a decoder of
its own, "Failed to decode X509 certificate: BER decoding error:
Expected Tag { class: Universal, value: 16 } tag…", which is a third
rejection wording and the reason the S/MIME half is real rather than
decorative. The mock now returns it for a certificate, so the branch
exists somewhere a client can meet it.
One thing found on the way: expiresAt is the registry's field and is not
derived from the key. A certificate valid for a year registers with
expiresAt null, so the card says "No expiry set" about a credential that
does expire. Left as it is, deliberately: reading the real date means
parsing the certificate, which is the second opinion this section
refuses to offer, and a date extracted here would disagree with the
server's own field the moment the two ever differed. What the row
reports is what the registry holds, and KNOWN-ISSUES says so.
A new Settings section, next to Identities & signatures: list, add,
rename and remove the OpenPGP public keys and S/MIME certificates
published on this account. Only public material -- no private key is
stored, requested or sent by any of this.
This is PR #67 revived. That branch was built against 0.16.19, closed
unmerged on 2026-08-26, and shares no ancestry with main after the email
scrub, so it is ported rather than rebased: the four files it added are
carried over, the three it edited are applied by hand, and everything it
claimed was re-probed against the live 0.16.20 on 2026-09-05. The i18n
work is new -- nine catalogues landed on 2026-08-31, after that branch
was written.
What the re-probe confirmed, unchanged from 0.16.19:
- An ordinary user may read *and* write their own keys, though the
permissions table lists every sysPublicKey* permission as
administrative. get and query both answered for a normal account,
and a malformed create came back invalidProperties naming `key`
rather than forbidden -- a rejection of the key, not of the person.
- The server parses the key and says precisely what is wrong. So
ihasmail does not validate key material; the server's message is
shown verbatim, as password-policy rejections already are.
- urn:stalwart:jmap is still absent from the session's top-level
capabilities and present per-account, so the check that reads all
three places is still the one that works.
What it added, none of which was known before:
- A key can parse perfectly and still be refused, with different
words: a sign-and-certify key with no encryption subkey -- what
`gpg --quick-generate-key` produces -- gets "Could not find any
suitable keys in OpenPGP public key". That is the rejection somebody
exporting from GnuPG will actually meet, and it is not a paste
error, so collapsing both to "invalid key" would send them back to
the clipboard for a problem that is in the key.
- emailAddresses comes back as {} when empty -- an object where a JMAP
list property should be an array. It type-checks, then throws in
join() while the list renders. normalize() checked the shape
already; there is now a test saying why, and the mock answers {} the
same way, because one that helpfully returned [] would let that
crash ship.
- A create answers with the id alone, no createdAt, so adding a key
reloads rather than believing the response.
- destroy works and leaves the registry empty. PR #67 shipped that
path untested -- its live probe was refused before anything was
created, so there was nothing to destroy.
- Patching `key` is allowed by the server. The mock still refuses it,
now deliberately rather than for want of evidence: ihasmail replaces
a key by adding one and removing the old, which keeps createdAt
meaning what it says.
x:EncryptionAtRest still does not exist on 0.16.20 -- asking for it is
an unknownMethod. encryptionAtRest is a field on x:AccountSettings, and
its value is a typed object ({"@type":"Disabled"}) rather than the bare
string ROADMAP described. Nothing here writes it.
An empty description is now sent as empty rather than filled in with
"Key". The description is stored on the server, so a default invented in
the client would be whichever language the adder happened to be using;
the list labels a blank one at render time instead.
Verified in a browser against the mock, not only in tests: both
rejections reach the toast in the server's own words with the form still
filled in, a good key renders its card, the kind is labelled from the
armour header, renaming persists, removing asks first and empties the
list, and the whole section reads correctly in German.
Raised on #174 as the other half of a migration -- import, notice
something is wrong, empty the book, correct the export, import again --
and tracked as #277.
The gap turned out to be wider than the ask. Contacts had no multi-select
at all: the only delete in the module was the cross on a single card's
pane, one card and one confirmation at a time. `destroyCards` has taken a
list and batched it against maxObjectsInSet since #218, and nothing in
the UI ever handed it more than one id. So "empty this address book" was
missing, and so was "delete these fourteen".
The list now has checkboxes, on hover the way the message list's are, and
always on a touchscreen where there is no hover to reveal them.
Shift-click takes the run between two rows. The search box gives way to a
selection bar rather than sitting beside it, because what the count
promises is what the search left on screen. A selection is cleared when
the book being shown changes, since carrying it across would leave a
count describing rows that are no longer there and a Delete aimed at
them.
Emptying a book is in the book's own menu, beside the import and export
that moved there in #226, and separate from Delete, which takes the book
with it. A default book cannot be deleted and can perfectly well be
emptied, which is most of the reason it is its own entry.
The part that is not a deletion, and the reason this is not one destroy
over everything in the book: a card filed in two books belongs to both,
and `ContactCard/set destroy` takes it away from both at once. Emptying
one book must not empty another, so a card with a second home is patched
out of this one and left alone. That is reported separately afterwards,
because it would otherwise look like contacts that refused to go.
`destroyCards` now answers with what the server confirmed rather than
throwing on the first refusal. A refusal that took half a selection with
it still deleted the other half, and an error saying only that it failed
sends somebody looking for contacts that are already gone. Both callers
report the count and the reason apart.
Emptying a shared book is deliberately not offered: the cards live in the
owner's account and this client has no path to write there.
One bug found by driving the built app rather than by any test, and worth
recording because of where it hid. The range a shift-click covers was
measured inside the `setPicked` updater -- which React runs when it gets
round to rendering, by which time the anchor ref has already been moved
to the row that *ended* the range. Every shift-click selected exactly one
row, and every store assertion still passed, because nothing was wrong
below the component. The anchor is read before the updater now, and the
contacts view has its first component tests: ten of them, six of which
fail if the measurement moves back inside.
Twelve new strings, in all nine catalogues, so nothing new falls back to
English.
Reported again by the submitter's colleague at LINET after #223 was
closed: duplicate checking was implemented for vCard and never for LDIF,
so re-importing an address book still leaves a second copy of everything.
That was deliberate at the time -- the matching key was an open question
I did not want to answer alone -- but the answer had already been given
on #174 and I closed the issue without acting on it.
The answer, in the submitter's words: an attribute that *can* change is
fine, because it will not have changed between two imports minutes apart.
An import is not a sync. That makes the `dn` usable -- it is the only
identity the file carries, and Mozilla's schema defines no UID -- and it
needs no guessing at all, unlike the name-plus-email fallback I had been
weighing.
So `uidFromDn` derives a namespaced, stable uid from the distinguished
name, normalised for the case and spacing two exports of one directory
differ in. A card the book already holds under that uid is updated rather
than duplicated, merged the way the vCard import merges: what the file
carries wins, what it does not mention is left alone. Reported as created
and updated, which is the pair that was asked for.
Three things worth knowing:
Matching is per address book, so two customer directories that each hold
a `cn=John Smith` stay two people as long as they are filed separately.
Imported into one book they would merge, which is the one way this can be
wrong and the reason the escape hatch is worth naming.
The look-alike count stays, and now means something narrower: entries
that `dn` matching could not catch -- one whose `dn` moved between
exports, and anything imported before there was a `dn` to match on. Those
are still only counted, never merged.
A file holding two entries under one `dn` is malformed, since a directory
cannot, and now becomes one card instead of two sharing an identity.
FEATURES gains the re-import behaviour for both formats; it documented
neither.
Dragging an event across the month grid wrote the date of the cell it
landed on into the event's stored start. Those are the same date only
while the event's time zone is the reader's.
An event kept in Asia/Tokyo at 15:00 is drawn to a reader in Phoenix at
23:00 the previous evening. Dropped on the 11th, it was written as the
11th in Tokyo -- which is the 10th on screen. It went where its own
calendar said rather than where the pointer did, one day short, every
time.
Moving by the difference between the two local days instead moves it
exactly as far as the hand did, and adding whole days to a stored wall
clock leaves the time of day alone without touching the zone -- so the
frame the rest of this path is careful about is still not crossed.
Found by giving the mock an event in a zone that is not the machine's.
Every other fixture used the machine's own, which cannot tell a correct
conversion from no conversion at all: the case that works is the one the
fixtures were all testing.
A JMAP push subscription stays silent until the client echoes back a
verification code. When the code arrives with no tab open, the service
worker leaves it in the cache for the next tab to collect.
Both sides named that entry relatively, and a relative key is resolved
against the URL of whoever is asking. The worker lives at <base>/sw.js, so
it wrote under <base>/; a tab at /mail/inbox/abc looked under
/mail/inbox/. They agreed only when the open page happened to be the root,
which is why this survived: the case that works is the one people try
first.
The failure is quiet in the worst way. A subscription that never gets its
code back simply never delivers, which is indistinguishable from push not
working at all -- there is no error anywhere to notice.
Both sides now build the key from the mount: the worker from the BASE it
already derives from its own location, the page through withBase. Found
while adding BASE_PATH, where the two disagree at every route rather than
only at deep ones; left alone then because it was pre-existing and
unrelated to that change.
FEATURES has always said attach-from-Files works "however large", because a
blob the account already holds is attached by reference and nothing is
sent. The code checked every file against maxSizeUpload regardless, so the
two disagreed and the code was the one that was wrong.
maxSizeUpload is what the server will accept for a single upload (RFC
8620). It bears on a file that is about to be uploaded and on nothing
else. Applying it to a by-reference attachment refused a 60 MB message the
server was already storing, on the grounds that it could not have been
uploaded -- which it was not being. Forwarding a large message as an
attachment hit exactly that.
A file from somebody else's account is fetched and re-uploaded into this
one, because a message can only carry blobs from the account sending it.
That upload is real and the limit is real for it, so it still applies
there.
A timetable, a rota, a public holiday list: the calendars people are given
as a link, which ihasmail could not show at all.
Nothing is stored. The document is fetched when the calendar is opened and
parsed in the browser; the server keeps no copy, no cache and no schedule,
which is what lets an immutable container serve this. There is no timer
either -- there is nowhere to run one -- so the guarantee is that a
subscription is as current as the last time somebody looked, which is also
when it matters. That is said plainly rather than implied.
The fetch has to happen on the server: a calendar URL belongs to whoever
published it and almost none of them send CORS headers. That makes it the
second place this app knocks on a door somebody else chose, so the guard
the image proxy has always had was lifted out and both now call it. A
second SSRF implementation is how one of them ends up missing a case; this
way there is one, and the extraction is covered by the image proxy's own
tests still passing unchanged.
webcal: is understood, because that is how these are published, and it is
read as https: rather than waved past the checks -- a webcal URL pointing
at loopback is refused exactly like an http one.
Recurrence is deliberately not expanded. RRULE is a small language with a
lot of edge cases, and a subscription quietly showing the wrong dates would
be worse than one showing the first occurrence and saying so.
The parser is a subscription parser rather than an importer: a subscribed
calendar is read-only and redrawn from scratch each refresh, so nothing has
to round-trip or survive an edit, which is most of what makes a full
iCalendar implementation large. What it does have to do is never mis-state
a time -- a DATE is built in local time rather than at UTC midnight, which
would land on the day before for anyone west of Greenwich -- and never hang
on a document somebody else wrote.
Events go through instancesIn like the birthdays, so no view has to know
they are not real calendars, and the calendar they hang off reports no
write rights, so everything that asks before offering an edit declines on
its own. A subscription that cannot be read says so in the sidebar rather
than drawing an empty calendar, which looks like a calendar with nothing
in it.
The calendar could only be edited through the editor, so moving a meeting
half an hour meant opening a dialog, changing two fields and saving. Every
other surface a finger or a pointer drives already answers to a drag.
In the day and week grids an event moves by dragging it and changes length
by dragging its bottom edge, snapping to fifteen minutes. In the month
grid it moves to another day and keeps the time it had, because a month
cell is a day and nothing finer -- changing the hour as well would answer
a question nobody asked.
It goes through the same path a menu edit takes. A recurring event is
asked which dates it means, and the answer runs through runScoped, so a
date the server will only change as part of a whole series offers that
rather than failing.
Three things do not offer a drag, and the reasons are checked in one place
so no grid has to remember all three: a read-only calendar, an event with
no calendar, and a birthday -- which is derived from a contact and has
nothing on the server to move. The reserved classes the swipe gesture was
told to keep out of are exactly the ones that are draggable here, which is
what that reservation was for.
Invitations are not sent. A drag is a scheduling gesture, and mailing
every guest on each nudge of a block is not what the hand was asking for;
a change that should go out with notice goes through the editor.
The new time is computed in the event's own frame rather than through an
instant. Working it out from the reader's local hours and then
re-expressing it in the event's zone converts twice, and the two do not
cancel: caught in the browser, where an event moved two hours the first
time it was dragged in the month grid and then sat still, because after
that its stored time and the reader's agreed. Parsing the stored string
into its parts and adding minutes to those touches no zone at all, and a
resize sends only a duration, so the question does not arise there either.
The theme was one enum -- system, light, dark, ihasmail -- where one value
carried a whole palette and implied dark. That works for exactly one
palette. The two questions now come apart: which palette, and which side.
Classic is the plain light and dark this app has always had. ihasmail's
own palette gains a day version, so the background of the dark one becomes
the text of the light one and the two read as one palette from either end.
Dracula, Gruvbox, Rosé Pine and Tokyo Night are the work of their own
projects, used under the MIT licence, and taken from each project's own
repository rather than from anyone's reimplementation. What was fetched is
recorded in .palette-sources/ and credited in NOTICE.
Giving ihasmail's palette a light half removed a whole special case.
Nothing is one-sided any more, so a palette can no longer override the
mode, the toggle no longer has to set a palette aside on the way to light
and remember it, and the greyed-out control that explained all that is
gone. The old lastDarkTheme, which existed only for that, is gone with it.
The shades between the published colours are derived rather than guessed:
these projects publish twelve to twenty values and ihasmail needs about
thirty. scripts/build-palettes.py computes the tiers and then measures
every text colour against the surface it sits on -- 4.5:1 for prose, 3:1
for borders and marks -- lifting anything short towards white on a dark
ground and towards black on a light one, so a lifted tier keeps its hue.
It refuses to write a palette that would not pass.
Every one of the nine halves needed at least one lift. These palettes are
built for code editors, not for prose at this size: Dracula's comment grey
is 3.03:1 on its own background and Rosé Pine's gold is 2.7:1 on Dawn.
Shipping them as published would have quietly ended the WCAG AA claim.
Two things caught while checking rather than while writing. The generated
blocks were appended to the end of the stylesheet, which put them after
the accent variants at equal specificity -- so choosing an accent over one
of the new palettes did nothing at all. They now sit before those rules,
where the existing ihasmail block's own comment says they have to. And
that block was unqualified, so it would have shadowed the new light half;
it is now explicitly the dark one.
Settings written before this carry `theme` and no palette, and are read
through the old enum. `theme` is still written back, derived, because a
device on an older build reads it and would otherwise be stranded on a
theme nobody chose.
Newest-first was the only order, so the mail you had not read yet was
wherever it happened to fall.
Seven presets and up to three levels of your own. It covers the Inbox
alone by default: unread-first is what people want in the folder they
triage and confusing in Sent, where everything is read and the order that
matters is when it went. Search keeps newest-first whatever the setting
says, since a result list is already ordered by the question that was
asked.
The server does the sorting, over the whole folder, for the same reason
search runs there: a list sorted in the browser is sorted only as far as
the browser has loaded, which on a folder of ten thousand is the first
fifty and a lie about the rest.
Two details that are easy to get wrong and were worth pinning in tests.
hasKeyword sorts a boolean and false comes before true, so "unread first"
is $seen ASCENDING while "starred first" is $flagged DESCENDING -- the
other way round. Getting either backwards puts exactly the mail you were
looking for at the bottom. And every order ends with newest-first as a
tiebreak, because a sort whose last level is a keyword or a subject leaves
every tie undefined, and an undefined order changes between two looks at
the same folder for no reason the reader can see.
Sorting on a keyword is optional in RFC 8621, and a server that will not
do it fails the whole query rather than degrading it -- so this setting
could turn a folder into one that does not open. The refusal is caught
once, the keyword levels dropped and the query retried, and nothing is
said: the reader asked for an order and got the closest the server can
give, and a toast on every folder change would be the app complaining
about its own request.
The mock now honours the sort instead of always answering newest-first,
which had it reproducing a server that silently returns a different order
from the one asked for -- the one shape of wrongness a client cannot
detect. MOCK_NO_KEYWORD_SORT=1 reproduces a server that refuses the
keyword sorts, so the fallback can be developed against.
The dates were already on the contact cards and nothing ever showed them,
so the one thing a birthday is for -- noticing it in time -- was the one
thing the app could not do with it.
Derived, not stored. The dates stay on the cards: a second copy of the
same fact drifts the first time somebody corrects one, and keeping a
calendar of its own is exactly what ihasmail does not do. Entries are
generated when a view asks for a range and vanish when the contact does.
They go through instancesIn like everything else, so no view has to know
they are different.
Off until switched on. It is derived data, and a calendar that fills
itself with dates nobody put there is a surprise rather than a feature. It
can also be hidden from the calendar's own sidebar without being turned
off, which is the same distinction the shared calendars already draw.
They cannot be edited or deleted, and that falls out of the design rather
than being special-cased: the virtual calendar reports no write rights, so
every control that already asks before offering Edit or Delete declines on
its own. updateEvent and destroyEvent refuse a synthesised id as well, so
the store is safe whatever calls it -- including anything added later.
Two things about the dates themselves. A card that records only a day and
month is the common case rather than the exceptional one, and gets a
birthday with no age rather than no birthday. And 29 February falls on the
28th in a year that has no 29th: somebody born in February has a birthday
in February, and moving it into March is the arithmetic winning over the
fact. Both are conventions; these are the ones that keep the fact intact.
The mock now carries birthdays on most of its contacts, including one with
no year and one on 29 February, so both cases are visible without a real
address book.
Both sides added a field next to the mail store's selection: the label
counts the sidebar draws, and the flag for a selection that means the
whole query rather than the loaded page. They are independent, so the
resolution keeps both.
Outlook sending in Rich Text packs every attachment into one TNEF blob.
Every other client shows a single unopenable winmail.dat, and the files
inside it are gone as far as the reader is concerned -- which is a
decoding problem rather than a mail one.
Written from the published format: a signature, a key, then a flat run of
attributes, each one a level byte, a 32-bit id carrying its own type, a
length, the data and a checksum. Attachments are delimited by
attAttachRenddata rather than named, which is why the parse is a small
state machine.
The MAPI property stream inside attAttachment is read for two properties:
the long filename and the MIME type. attAttachTitle carries an 8.3 name,
so a file that arrived as "Quarterly Report Final.docx" is QUARTE~1.DOC
there and correct here. The stream stops at a named property (id >=
0x8000) rather than guessing past it, since those carry a GUID before
their value and nothing after one can be trusted to stay aligned.
Decoded in the browser, on request. The server never sees the contents and
has nowhere to keep a decoded copy; doing the work on sight would spend
the bandwidth whether or not anybody wanted what is inside.
A blob that goes wrong part-way through keeps what was read before that
point, whether it ran out or the checksum stopped matching. Half the
attachments beats none: the alternative is a reader who can see the file
is there and cannot have it. The original stays attached either way.
The message body is deliberately not decoded. TNEF can also carry it as
compressed RTF, which is a second format again for a body the reader
already has in plain text or HTML nine times in ten.
The mock now sends one, built by its own encoder rather than by the
parser's fixtures, so the two are independent implementations of the same
description.
A flat list is fine at five labels and unreadable at thirty, and there was
no way to keep one that matters occasionally without it holding a row for
ever.
A label can now sit under another, and each says whether it belongs in the
sidebar always, only while it has unread mail, or never.
Nesting is display only. The keywords stay flat on the message, which is
what keeps them readable by every other client: moving a label under
another rewrites nothing in the mailbox, and a client that knows nothing
about ihasmail sees exactly what it always did. Both new fields are
optional, so a settings file written before this parses unchanged and
means what it did.
Settings sync between devices, so the tree has to survive shapes that
should not exist. A label whose parent was deleted on another device comes
back to the top level rather than vanishing -- a label that disappears
because something else was deleted is one the reader cannot get back. A
cycle arriving from an older device is broken by treating the label that
closes the loop as a root, so nothing is lost and nothing hangs. The
parent picker will not offer a label's own descendants, so one cannot be
built here in the first place.
A label kept by the unread rule keeps its ancestors, whatever they were
set to. A child cannot be drawn under a parent that is not there, and
promoting it to the top level would silently rearrange the tree at the
moment the reader is least able to explain why. The parent comes back as a
container instead, and its own count still says whether it has anything of
its own.
Unread counts come from one request carrying a query per label rather than
a request each, with limit 0 so the server does not send ids that would
only be thrown away. They refresh on the same beat as the folder counts,
since the things that move them are the same things, and a failure is
swallowed: a count is decoration, and the sidebar draws the label without
one.
Also corrects the Labels page, which said names and colours are kept in
this browser. They live in the account's own Files and follow it between
devices, like every other setting that is not about this screen.
The header checkbox selected the loaded page. On a folder of ten thousand
that is fifty of them, and the only way to act on the rest was to scroll
until they loaded and tick again.
A line now offers the rest by name once the page is selected, and taking
it is a separate press. A checkbox that silently meant ten thousand when
the screen shows fifty would be the worst of both, so each option says
what it actually covers.
The wider selection is a query rather than a list of ids. What it reaches
is resolved from the server when an action runs, walked a page at a time,
because a folder holds far more than one call returns and Email/set
refuses more ids than maxObjectsInSet in one go -- which setEmails already
chunks for. It resolves uncollapsed, unlike the list: "everything in this
folder" means every message rather than one per thread, and expanding
threads the way a click does is impossible here anyway, since that walks
loaded Email objects and these are the ones that were never loaded.
Two things this exposed.
Undo is now withheld once a move reaches messages that were never loaded.
It restores the folders each message was in, taken from what the browser
holds, and for an unloaded message that is nothing -- so the undo would
have written an empty mailboxIds and left the message in no folder at all,
which is worse than the move it was undoing. move() and archiveByDate()
both did this; both now drop the offer rather than restore something
wrong.
And an action consumes the wider selection. The optimistic paths cleared
the selected ids but not the flag, so the next action would have silently
reached the whole folder again.
Four warnings, in Privacy & safety, and all of them start switched off.
That is not timidity. A client that begins by interrupting is one people
learn to click through, and a warning clicked through without reading
costs the same attention and buys nothing. The outside-sender warning
could not be on by default in any case: it measures against the domains
that count as yours, and with nothing configured every message in the
mailbox is from outside.
Your own identity domains are always internal and are not configuration.
An account signed in as [email protected] warning that example.com is
external would be absurd, and making it be typed in first is a foot-gun
that leaves the feature useless the moment it is enabled. Configured
domains are additional, and cover their subdomains -- matched on a dot
boundary, so example.com covers mail.example.com and not notexample.com,
which is exactly the domain somebody registers on purpose.
The four:
A banner names the sender's domain on a message from outside.
Sending outside names the outside recipients and asks, rather than
refusing. "This is going outside" is a rule and not something the sender
can check; a list of addresses is. It reads the full identity list rather
than the visible one, since hiding an identity from the From menu does not
make its domain somebody else's.
Sending to a large group asks once the count crosses a threshold, which is
what catches a reply-all onto a long thread. It counts people rather than
headers, so one address in To and nine in Cc is a message to ten.
Opening a link asks when the destination is not trusted, and always when
the link's text names one domain while its destination is another -- even
where that destination is trusted, because being trusted is not the same
as being the place the text claimed. On that mismatch the offer to trust
the domain is withheld: what would be trusted is the destination, and the
destination is not the thing in question. Anything that is not http or
https is left alone, since warning about a mailto: is noise and noise is
how a warning stops being read. Both bodies are covered, because a link in
a plain-text mail is linkified by us and points wherever it likes just as
readily as one the sender marked up.
The click is cancelled and the navigation re-issued after the answer,
since there is no way to hold a real navigation open across a dialog. The
reopen runs in the continuation of the dialog's own click, which is still
the gesture a popup blocker wants to see.
`BASE_PATH=/mail` mounts the whole app under a prefix, for a host that is not
ihasmail's alone. Unset -- every deployment that exists -- is the domain root
and is byte-for-byte what it was: the canonical form of the setting is the
empty string, and `""` concatenated onto `/api/health` is `/api/health`.
That choice of canonical form is the whole design. A trailing slash would have
been the obvious alternative, and it fails quietly in exactly one place: at the
root it makes `//api/health`, which is not a path on this host but a
protocol-relative URL to a host called `api`. One call site forgetting to
branch is a request leaving the origin. So the empty string, one leading slash,
no trailing one, worked out once in `scripts/basePath.mjs` -- plain JS, next to
`version.mjs`, because the web build and the server both have to reach the same
answer and two implementations of "what does /mail/ mean" is precisely the bug
where the server serves an app whose script tags point somewhere else.
`/mail`, `mail`, `/mail/` and `//mail//` all mean the same mount; a deployment
should not fail over a trailing slash.
Unlike everything else ihasmail is told, this one cannot wait for the process
to start. The bundle writes its own asset URLs into index.html, so `BASE_PATH`
is read at build time for Vite's `base` as well as at run time for the routes,
and the Dockerfile carries one value into both. Get them out of step and the
page comes up blank with a 404 in a console nobody has open -- so the static
handler, which is reading index.html anyway, checks what it asks for and says
so in the log once per build.
Everything moves together. The API mounts at `${base}/api`; the router is
given the base once, so every `<Route path>` and `<Link href>` stays written
root-absolute and wouter does the rest; `apiFetch` adds the prefix in one place
rather than at forty call sites; the session cookie's Path narrows to the mount
so two instances on one host cannot sign each other out.
Two things need no prefix at all, and it is worth saying why they were not
given one. A manifest's members resolve against the manifest's own address, so
relative URLs there follow the mount with nothing substituted at build time --
which is also why `public/` needed no template step. The service worker is the
same trick: it is served from the mount, so `new URL("./", self.location)`
tells it where that is, and a worker that derives the value cannot disagree
with the page that registered it.
Anything outside the mount is a 404 rather than the app shell, and
`stripBasePath` does not use `startsWith` -- under `/mail` this process shares
a hostname, and answering `/mailbox` with our index would shadow a neighbour
instead of letting it 404 honestly. For the same reason the notification-click
handler now checks the path as well as the origin: `includeUncontrolled` widens
`matchAll` to the whole origin, which off the root would have navigated a
stranger's tab to our inbox.
Inline images in a draft were the one silent trap. They are matched by their
blob URL on the way out, once unanchored and once anchored, and a bare
`/api/blob/` still appears inside `/mail/api/blob/...` -- so one pattern would
have replaced the tail and left `/mail` in front of a `cid:`, and the other
would have missed and sent the message linking to the sender's own webmail.
Both patterns are built from the base now.
General had grown five unrelated headings and was where anything without
an obvious home ended up. Remote images were filed under "Reading", the
read-receipt policy under "Composing", the undo-send window beside the
default message format. They are the same kind of decision -- what reaches
a sender, and what asks before something happens -- and they were the
hardest settings in the app to find.
Privacy & safety now holds all six, in three groups: remote content, read
receipts, and the things that ask before it is too late. General keeps
what it is actually about and is thirty lines shorter.
The line against Security & sessions is worth stating, because two similar
words next to each other in a nav is how a menu becomes something people
hunt through. Security & sessions is credentials and access: password,
two-factor state, app passwords, live sessions. Privacy & safety is how
the app behaves towards the reader and towards senders.
Nothing moved in storage. Settings are a flat object in settings.json and
sections are only how they are grouped on screen, so this is a UI change
with no migration and no key renames.
Two things beyond the move.
The senders trusted with remote images are now listed and can be withdrawn
one at a time. A sender was added from a message and could then only be
removed by finding another message from that same sender, which is not a
way to review a list you cannot see.
And General's lead said settings are stored in this browser, which is only
true when the server has no FileNode support. They normally live in the
account's own Files and follow it between devices, so the sentence
contradicted the feature it sat above.
The calendar had next and previous as toolbar buttons and n/p, and nothing
for a thumb. Every other surface in the app that a finger drives already
answers to a swipe.
Day and month only. Those are the two views where a period is a page;
week and agenda scroll through a range rather than turning to the next
one, so there is nothing a sideways flick would obviously mean. Dragging
left pulls the next period in from the right, the way paper and every
phone do it.
Three things it deliberately does not do.
It draws nothing while the finger moves. The row swipe slides the row open
because the strip underneath has to name which of six actions is about to
happen; stepping a calendar has two outcomes and the direction of the
finger already says which. Translating the grid would also break the
sticky day header, since a transform makes a containing block. The
threshold is reported by the vibration motor instead, which is what the
haptics are for.
It does not start on an event. Which gesture was meant is decidable at the
moment the finger lands and only then, so dragging an event stays
available to be built later without having to be untangled from this
first.
And it asks for a longer drag than a row swipe -- not because the
consequence is bigger, since stepping back undoes it while a swiped row
has already been archived, but because this gesture has no way to change
its mind. A row reveals what it will do and can be let go early, and
offers Undo after. This shows nothing and offers nothing, so the distance
is the only chance to not mean it.
The axis lock is the shared one, keeping its bias towards the vertical:
the day grid scrolls through the hours, and a scroll misread as a swipe
throws the reader into another day. The toolbar buttons and n/p stay,
because a gesture with no visible control is one only the people who
already know about it can use.
Archiving put everything in one folder, so an Archive that has been
collecting for years is a single flat list with no way to narrow it except
search.
Archive by year and Archive by month file into Archive/2026 and
Archive/2026/09, creating the folders as needed and reusing them after
that, including folders made by hand or by another client.
The names are numeric and zero-padded rather than month names, because
these are real server-side mailboxes rather than anything of ihasmail's.
Every other client sees them: a folder created as "September" by someone
reading in English stays "September" for the same account read in
Japanese, since the name is stored and not translated. And 09 sorts
between 08 and 10 where a name does not.
The date is read in the reader's own timezone rather than UTC so it agrees
with the date shown against the message in the list. A message that
arrived at 00:30 UTC on 1 September is dated 31 August in New York, and
filing it under 09 while the list says August would be the app
disagreeing with itself. A message whose date cannot be read goes to
Archive itself rather than to a folder named after a guess.
A selection spanning two months is two destinations, not one. The moves
are made silently and one toast names where everything went -- the folder
where there is a single answer, the count where there is not -- because
each group raising its own toast with its own Undo would mean undoing a
third of a move. One Undo restores the whole selection to wherever each
message came from, captured before anything moved.
The menu labels name the destination where there is one, so it reads
"Archive to 2026/09" rather than describing the rule, and falls back to
"Archive by month" for a selection with no single answer.
Forwarding quoted the original into a new message, which is the right
thing for passing on something to be read and the wrong thing for passing
on something to be looked at. Quoting rewrites the body, drops the
headers, and re-parents the attachments, so a bounce, a phishing report or
anything else where the message itself is the evidence arrived altered.
Forward as attachment sends the message whole, as a message/rfc822 part.
It costs no upload at all: a message's own blobId is its RFC822 blob and
already lives in the account, so this goes through the same by-reference
path as attach-from-Files and a 40 MB message attaches as fast as a small
one.
It is in the message's own menu, the list's right-click menu, and the
overflow on the reply strip at the foot of a thread, which is the one a
thumb finds on a phone.
Two things fixed on the way, both exposed rather than introduced by this.
The filename rule was subject.replace(/[^\w.-]+/g, "_"), and \w without
the u flag is ASCII: every character of a Russian, Japanese or Chinese
subject failed the class, so those messages downloaded as a row of
underscores. What is actually unsafe in a filename is much shorter than
"not ASCII" -- path separators, the names Windows reserves, the control
range -- so the rule now keeps letters from any script and drops only
those. It lives in one place and the .eml download uses it too.
And the composer's attachment chip set overflow/text-overflow on a span,
where neither does anything, so the name never truncated and the size ran
on after it on the same line. Only long names showed it, which is every
.eml named from a subject.
The filter in front of the mailbox scores every delivered message and
writes its working into headers, and none of it was being read. A message
in Junk gave no reason for being there.
Nothing here scores anything. The headers are parsed and shown, so this
cannot disagree with the filter that actually made the decision.
Two formats cover what sits in front of a Stalwart mailbox in practice:
the SpamAssassin-shaped X-Spam-* set, which Stalwart's own filter writes,
and Rspamd's X-Spamd-Result. A header in neither shape is left unread
rather than guessed at, since a misparsed score shown confidently is worse
than no panel at all. Mail that arrived without any of them shows nothing.
Rules are listed largest mover first and signed, because which way a rule
pushed is the point, and the biggest contributor is the answer to why the
message scored what it did.
Two things it deliberately will not do. A score is always given the
threshold it was measured against, because 6.7 is damning against 5 and
unremarkable against 15 -- the number alone is not something a reader can
act on; where no threshold was stated, it says so rather than implying
one. And where the filter recorded no verdict, none is derived from score
against threshold: the filter applies policy we cannot see, and putting a
verdict in its mouth would be inventing one.
The mock writes the same headers at delivery -- spam in Junk, clean in the
Inbox, nothing on mail this account wrote -- so the panel can be developed
and demoed against it.
Templates were a fixed subject and body, so anything that changed per
message -- who it is going to, today's date -- had to be typed over
afterwards.
Eight names are recognised: recipientName, recipientFirstName,
recipientEmail, myName, myEmail, subject, date and time. Dates and times
go through datetime.ts rather than toLocaleDateString, so a template
follows the date order and clock the app was already told to use.
Filling happens on insert rather than on send. What a placeholder came to
is then visible in the composer and can be edited, instead of the message
changing between writing it and sending it.
Two things are deliberately left alone. A placeholder that cannot be
answered yet -- a recipient's name on a draft nobody has addressed -- stays
in the body as written, because substituting an empty string produces
"Hi ,", which is wrong rather than visibly unfinished; leaving the name
says which word is still missing and can be typed over. And a name that is
not a placeholder is left as written too, since a body that quietly ate an
unrecognised token would be worse than one that shows it.
Values are escaped on the way into HTML: a display name comes from a
contact card or a typed address and is not trusted markup.
FEATURES.md had nothing on three of them. The scheduling panel and what it
will not show, the filter editors no longer discarding work in silence,
and where compose-as-new can be reached from — including the reply strip,
which is the one a thumb finds.
KNOWN-ISSUES.md gains the free/busy finding, which is the sort of thing
that page exists for: it was checked against the live server rather than
assumed, free/busy between accounts turns out to need no sharing set up,
and a principal offers no route to its calendars at all. Also what the
check did *not* settle, and which way ihasmail errs in the meantime.
ROADMAP.md gains the two things left deliberately unbuilt: a scheduling
view you can visit with no event in hand, and per-message actions from the
message list on a touchscreen, where the gesture that would open them
already means "select".
Somebody arriving from SOGo, Thunderbird or an LDAP directory has their
contacts in LDIF, and until now the only way in was vCard.
Nothing on the server reads LDIF, so this reads it here, in two pieces
that are two different problems. `ldif.ts` is RFC 2849 and nothing else:
folded lines, base64 values, case-insensitive attribute names, options,
comments, `version:` headers, change records. It knows no attribute by
name. `mozillaAb.ts` knows the attributes and no syntax -- Mozilla's
address book schema, which is what Thunderbird and SOGo write and what the
issue asks for by name. LDIF says nothing about what any attribute means,
so a file is only readable against a schema, and keeping the two apart is
what would let a second schema be added without touching the reader.
Work and home addresses, which the schema keeps in two separate sets of
attributes, come across as two addresses. So do every phone kind, the
second email, the organisation and its units, job title, nickname, web
pages and the AIM handle. The four custom fields have no equivalent in
JSContact and are appended to the note, labelled as Thunderbird labels
them: keeping something somebody chose to write down is worth more than
the tidiness of dropping it.
An entry with neither a name nor an address is skipped rather than
imported as a blank row that is impossible to identify and tedious to find
again to delete. The distinguished name is not used as the contact's uid:
it says where an entry sat in somebody else's directory.
One import control takes either format and decides by what is in the file
rather than by what it is called, because an address book exported as LDIF
arrives as .ldif, .ldi, .txt or with no extension at all.
Closes#174
A mail the far end rejected, or one that went to an address with a typo in
it, is a mail you want to send again -- not forward, and not reply to.
Doing it by hand meant a new message and copying five fields across.
"Compose as new" sits with Reply and Forward, in the message menu and in
the list's right-click menu. It opens a composer holding the recipients
the message had, its Reply-To, its subject with nothing prefixed to it,
its body with nothing wrapped around it, and its attachments.
What makes it a new mail is what it leaves behind. `draftId` stays null,
or sending would destroy the message it was made from. `inReplyTo`,
`references`, `relatedEmailId` and `relatedKeyword` stay null, so it hangs
off no thread and sending it marks the original neither answered nor
forwarded. The Message-ID is the server's and the date is stamped at
build time, so both are new without anything asking for them -- the send
path needed no changes at all.
A message you sent is composed as the identity you sent it as. One
somebody else sent has no identity of yours to match, and guessing from
whom it was addressed to would put the resend behind an alias that was
only ever the receiving end, so that case takes the account's default.
No signature is added. The body is the one that was sent, which already
ends in whatever signature went with it, and appending the identity's
would give it two.
Closes#176
An .ics reaches you by ways that are not your mailbox -- a ticketing
system a customer invited, a colleague's export, a booking confirmation
forwarded on -- and until now the only events ihasmail could take were the
ones attached to a message it had received.
The calendar's own menu now offers "Import iCAL file…", which files
everything in the file into that calendar. No global button: the issue is
right that this is not a frequent enough thing to earn one.
The parsing is the server's, through the same CalendarEvent/parse an
emailed invitation already goes through. An .ics is not a format worth
reimplementing in a browser, and Stalwart's reader handles what a
hand-rolled one would not.
Every event goes out in a single CalendarEvent/set. The round trips are
the smaller half of the reason: createEvent invalidates on the way out and
invalidating refetches every cached range, so a year of events imported
one at a time would refetch the calendar a few hundred times.
Nothing is mailed to anyone named in the file. Importing is filing
something you already have, and scheduling messages would be a surprise to
its participants.
The mock's parser read the whole file with one regex and returned one
event, which is all an invitation ever needed. It now reads per VEVENT, so
a multi-event file can be tested against it, and it invents an organiser
and an attendee only for events that carry a METHOD -- a plain export is
not addressed to anyone.
Closes#173
Follow-up to #167: an event made from a mail now opens with the sender
and everyone it was addressed to already in the guest list, so a thread
becomes a meeting without retyping the room.
Two things are deliberately left out. The reader's own addresses, since
they are the organiser and an organiser among their own guests is an
invitation to your own appointment. And a blind copy, on a message the
reader sent: a guest list is visible to every guest, so promoting a Bcc
to a guest would tell the room about a copy the sender chose to hide.
That is not something a menu item may do quietly.
*Send invitation emails to guests* now starts off when the guests were
inherited rather than typed, and on everywhere else -- which is every
event whose guests somebody chose one at a time. The reason is the case
the issue opened with: a reminder made out of a bill carries the biller
and everyone else on the mail. Left on, the primary button reads Send
invites and the first press mails all of them an invitation to what was
meant as a note to self. The switch sits right there under the list and
says what it does, so inviting them is one deliberate click. Un-sending
is not.
Asked for in #167: a right-click on a mail that turns it into a calendar
entry, the way a bill or a task becomes a reminder. Nothing clever, and
deliberately so -- the subject becomes the title, the body becomes the
description, and the reader supplies the one thing the message cannot.
A due date is exactly that thing. "Due on the 14th" in an invoice is not
a date a parser could be trusted with, and a wrong guess quietly
scheduled is worse than no guess at all, so the editor opens on the next
half hour for an hour and the reader fixes it. Forward rather than now,
because a start time that has already passed by the time they press
Create is one more thing to correct.
The body is capped at 5000 characters. A newsletter is a message too,
and its whole body would be stored on the event, synced to every device,
and shown in a three-row textarea; what is worth keeping -- the amount,
the account, the address -- is near the top. The cut is marked, so a
truncated bill is not read as the whole of it.
One message only. The list menu acts on the selection everywhere else,
but there is no sensible event to make out of five mails, and the mobile
entry appears only when exactly one row is held.
The editor lives inside CalendarView and the reader is in the mail view
when they ask, so the draft waits in the calendar store until that view
mounts and takes it -- once, or it would reopen on every later visit.
It seeds a form rather than an event: the dialog still says New event
and still has to be pressed.
Called *Create event…* rather than "appointment", which is the word the
issue used: it opens the New event dialog, and each catalogue already
has its own settled noun for that -- Termin, événement, 日程.
Reachable three ways, since a phone has no right-click: the row context
menu, a message's ⋮, and the ⋮ of a held row on mobile. Hidden entirely
where the account has no calendar.
The picker offers English plus nine translations. I wrote it up as nine
in total with eight unread, which is off by one in the direction that
undercounts the work and, worse, misstates how many catalogues are
waiting for a speaker to read them.
Nine of the ten are machine-made and unread — all of them, not all but
one — so the sentence that matters reads more sharply than the wrong
version did, not less.
The translations shipped today and the docs still said "no languages but
English". They also need to say the harder thing, which is that eight of
the nine have never been read by anybody who speaks them.
- README gains the language list, with the Beta caveat in the same line
rather than a footnote.
- FEATURES.md gains an Interface language section: the list, why it is a
separate setting from the date locale, and the two design properties
that follow — a missing entry renders English, and plurals are asked of
Intl.PluralRules rather than assumed, which is why Russian carries
three forms and Japanese one.
- ROADMAP.md no longer lists translations as "not yet". What replaces it
is the half that is genuinely not done: a translation anybody has
checked. RTL is split out as its own entry, because holding Arabic,
Hebrew and Persian back is a layout decision and not a queue position.
- KNOWN-ISSUES.md gains two entries. The unread catalogues, which is the
one thing on that page that cannot be closed by testing. And the
coverage number that read 100% while two hundred strings rendered
English in every language — recorded as a general lesson rather than an
i18n one, since a coverage number measures what it can see and the rest
is exactly what nobody is checking.
Background notifications were built, verified against a live server, and then
went quiet a few days later on every device that had them. A JMAP push
subscription expires -- seven days is the ceiling -- and re-registering before
it lapses is the client's job. Nothing did: enableWebPush() was reachable only
from the switch in Settings, so the subscription was registered once, expired,
and stayed expired. Nobody reports that as a bug. They report that push does
not really work.
It is renewed on every app start now, which is the only place it can be: the
registration is a JMAP call and the service worker has no session cookie to
make one with. So the guarantee is that push keeps working as long as ihasmail
is opened now and again, and a two-day renewal window against a seven-day
ceiling means once a week is enough. Registering is the same call as turning it
on -- deviceClientId makes a repeat replace rather than accumulate -- so there
is no second path to get wrong.
Two more things in the same area, both of which produce the same silence:
- webPushActive() asked whether the *account* had any subscription, so the
moment one device had one, every other device showed the switch already on.
A phone that had never successfully registered, or whose registration had
since expired, read as on and delivered nothing. It matches on the device now.
- Turning push on reused an existing browser subscription and gave up if there
was none. A browser drops or rotates one on its own, and there is no tab open
to hear the pushsubscriptionchange when it does, so that state was permanent.
Renewal re-subscribes rather than bailing.
Whether this browser has push on is now remembered locally, which is what
renewal keys off. It is per browser rather than per account on purpose: a
subscription is an endpoint and a device, and a phone having push says nothing
about the desktop. It is not kept across sign-out, matching sign-out already
destroying the subscription itself.
The mock is the reason this was invisible in development: it handed back
expires: null, so a client that never renewed worked perfectly against it
forever. It expires a subscription in seven days now, which is what makes
"does this client renew?" a question the mock can answer.
Checked against the mock: a create returns an expiry seven days out that
survives PushSubscription/get and parses, renewing the same deviceClientId
replaces rather than accumulates, and a device with no registration of its own
finds nothing where the old code saw two subscriptions and said yes. What the
live Stalwart sets for expires is not confirmed -- if it sets none, renewal
correctly does nothing and the other two fixes still stand.
FEATURES.md and the touch gestures were written at the same time on separate
branches, so the inventory of everything ihasmail does landed knowing about
none of them: the mobile layout was still one line about a tab bar, and the
Appearance row of the settings table did not mention the two settings that
had just been added to it.
The gestures get a section of their own under Layout rather than a bullet,
because the interesting part is not the list of five -- it is why they are
gated on `(pointer: coarse)` rather than on width, why a direction with no
meaning in this folder refuses to move rather than moving and doing nothing,
and why the axis lock gives a diagonal drag to the scroller. The same
reasoning is in the code; this is where somebody reads it without opening
lib/touch.ts.
Also here: the `dvh` and safe-area notes alongside the mobile layout, a long
press added to the list's multi-select bullet, and a short note in the
settings-sync section on why the swipe actions follow the account despite
looking like a per-device setting.
The feature list lived in three places that each answered a different question: the site sells it, the docs teach the parts that surprise people, and the README summarises both in a paragraph. Nothing said, in one place and at full detail, what is actually built -- so evaluating ihasmail meant reading the source, and a feature that quietly stopped working had nowhere to be contradicted.
FEATURES.md is that inventory, written from the code rather than the copy: the capability matrix and what each missing one removes, the search operators as the parser actually reads them, the Sieve tests and actions, every settings section and which of them follow the account, the shortcut set, the security posture, and the environment. Where a behaviour is odd it says why, because the reason is usually the stateless constraint.
It also records what the docs had not caught up with: single-occurrence edit and delete are built, and the id renumbering that made them hard is described where someone changing that code will find it.
jcoffey-dev
Coffey Labs