We don't mount /boot in fstab to prevent systemd-udev looking for
corresponding kernel modules in /lib/modules. We need to do this
because the modified kernel-qemu image we use for emulation is usually
not 100% the same version as in the Raspbian image. Therefore
systemd-udev won't be able to find the 'correct' kernel modules in
/lib/modules.
This causes issues when the raspberrypi-kernel package is updated
however. apt will update /lib/modules and will write the new
kernel{7,}.img files to /boot. Before this patch /boot was just a
directory on /dev/sda2, not the 'real' /boot which is on /dev/sda1. So
you'd end up with old, not updated kernel images but new, updated
kernel modules in /lib/modules.
This mounts /boot after the system booted. /boot gets mapped to
/dev/sda1 by /etc/udev/rules.d/90-qemu.rules.
Fixes#12650
opw-682320
9f0120d73b solved this issue already, but
it was only applied to POS orderlines. There is other data on the
report (eg. payments, taxes, ...) that should also correctly respect the
user time zone as well.
opw-678748
Chrome 50 treats percent-height divs inside of auto-height cells as
auto [1]. So from now on it's important that an explicit 'height: 100%' CSS
property is set on parent tds, otherwise you'll end up with elements
with a height of 0.
An extra difficulty is that this new height property on
subwindow-container will result in the element being as high as his
parent table. So the collapsed trick doesn't work anymore in the
customer list.
This has to be done conditionally. The proposed workaround of adding
100% height to parents of affected elements causes issues in IE/Edge
because the effect of adding a height in percent to a table-{cell,row}
element is not defined by CSS [2].
DO NOT FORWARD-PORT!
[1] 8876584335
[2] http://stackoverflow.com/a/27384730
Chrome 50 treats percent-height divs inside of auto-height cells as
auto [1]. So from now on it's important that an explicit 'height: 100%' CSS
property is set on parent tds, otherwise you'll end up with elements
with a height of 0.
DO NOT FORWARD-PORT!
[1] 8876584335
Catching and hiding database transactional errors can
sometimes cause a POS order to be entirely lost.
When it occurs, the transaction won't be committed
into the database, and if there is only one order
in the batch, the server won't return any error to
the frontend POS which will consider the order saved.
When the POSBox boots without a network cable attached it will
automatically launch a wireless AP that people can connect to. This
allows them to configure what wireless network the POSBox should connect
to.
This wireless AP was configured to use the 10.10.0.0/24 subnet. The AP
itself was on 10.10.0.1. Although this is fine if used as intended it is
a quick way to take down an existing network if you where to plug in an
ethernet cable after the wireless AP has started. 10.10.0.1 is commonly
used by routers all over the world and plugging in a booted POSBox into
their networks will cause serious issues because the POSBox will share
the same IP as the router.
This moves the POSBox AP to the 10.11.12.0/24 subnet, with the AP on
10.11.12.1.
This also makes the DHCP server listen only on wlan0 because otherwise
you can end up with two DHCP servers on the same network which would
still break stuff.
Useful when updates to the initialization scripts don't go as
planned. This leaves something to inspect.
The initialization script already automatically stops (because of 'set
-o errexit') but it was a bit tricky to actually see what went wrong
because scrollback in QEMU isn't great.
The main reason for doing this is supporting the new Raspberry Pi 3. No
functional changes where made.
For Raspbian Wheezy we used to download the full image and strip it as
best we could to obtain a reasonable image size for people to
download. Since Raspbian Jessie the Raspberry Pi Foundation has started
releasing an official minimal image (Raspbian Jessie Lite) which we will
use from now on to build our image. One downside of this is that the
minimal image is a 1.3 GiB image which is too small for our
purposes so it has to be resized.
Because Raspbian Jessie migrated to systemd we cannot rely on
/etc/init.d/rcS to set up the ramdisks anymore. Jessie provides a
compatibility layer so old SysVinit scripts still work but rcS does not
block like it does in a SysVinit system, it is run in parallel with
other startup services. In our case this is a bad thing as setting up
the ramdisks has to be done before any other services are started. To
accomplish this the rcS hack has been migrated to a systemd service
running before basic.target and with DefaultDependencies=no. This has a
similar effect as the rcS hack because normal systemd services (with
DefaultDependencies=yes) all require basic.target by default.
In v8.0 and saas-6, the PoS does not support the concept of fiscal
position. Therefore, the invoice generated should not use it, otherwise
there might be an inconsistency between the PoS order and the invoice.
FORWARDPORT TO SAAS-6 ONLY!
Fixes#11299
opw-671743
Edge >=12 throws an error (code 800a025e) when calling the select()
method on an input element that is not currently part of document and
has a non-empty value.
When adding a paymentline a "change:selected_paymentline" event gets
triggered which in turn triggers the focus_selected_line function before
the actual paymentline gets rendered.
opw-671426
When we render the payment screen the current value of the current
paymentline is rendered to a string using
Paymentline.get_amount_str(). This conversion always rendered numbers
with a decimal separator of '.' because it just used
Number.toFixed(). This used to be fine but
5a10903e9b made the POS adhere to the set
decimal and thousands separator. Because of this, the POS now has to
parse the input with web.parse_value() which takes into account these
separators. When parse_value() parses a float it first deletes the
thousands separators and then replaces the decimal separator with '.'
before it gets cast to a Number.
In German (and other languages), the '.' is used as a thousands
separator which means that it will get deleted by parse_value() so
you'll end up with '12.34' being interpreted as '1234'. Because of this
it is important that we ensure that the default value that appears in
the input uses the correct separators.
The aforementioned issue only occurs when the user changes the default
value in the paymentline. Only then does a conversion take place from
String to Number. So just clicking on a payment method of type 'Bank'
and validating it would work fine, only when the payment amount was
updated would the issue occur.
Fixes#9395Closes#10952
note: this should not be forward-ported to >8.0 because later versions
use the rewritten payment screen which is completely different from the
one in 8.0.