Enabling the following features for all targets, except for nrf51
targets compiled to be used with SoftDevice:
- MICROPY_PY_ARRAY_SLICE_ASSIGN
- MICROPY_PY_SYS_STDFILES
- MICROPY_PY_UBINASCII
Splitting mpconfigport.h into multiple device specific
files in order to facilitate variations between devices.
Due to the fact that the devices might have variations in
features and also variations in flash size it makes sense
that some devices offers more functionality than others
without being limited by restricted devices.
For example more micropython features can be activated for
nrf52840 with 1MB flash, compared to nrf51 with 256KB.
We're moving towards a co-processor model and a Wiznet library is
already available.
New native APIs will replace these for chips with networking like the
ESP32S2 but they won't be these.
The motivation for doing this is so that we can allow
common_hal_mcu_disable_interrupts in IRQ context, something that works
on other ports, but not on nRF with SD enabled. This is because
when SD is enabled, calling sd_softdevice_is_enabled in the context
of an interrupt with priority 2 or 3 causes a HardFault. We have chosen
to give the USB interrupt priority 2 on nRF, the highest priority that
is compatible with SD.
Since at least SoftDevice s130 v2.0.1, sd_nvic_critical_region_enter/exit
have been implemented as inline functions and are safe to call even if
softdevice is not enabled. Reference kindly provided by danh:
https://devzone.nordicsemi.com/f/nordic-q-a/29553/sd_nvic_critical_region_enter-exit-missing-in-s130-v2
Switching to these as the default/only way to enable/disable interrupts
simplifies things, and fixes several problems and potential problems:
* Interrupts at priority 2 or 3 could not call common_hal_mcu_disable_interrupts
because the call to sd_softdevice_is_enabled would HardFault
* Hypothetically, the state of sd_softdevice_is_enabled
could change from the disable to the enable call, meaning the calls
would not match (__disable_irq() could be balanced with
sd_nvic_critical_region_exit).
This also fixes a problem I believe would exist if disable() were called
twice when SD is enabled. There is a single "is_nested_critical_region"
flag, and the second call would set it to 1. Both of the enable()
calls that followed would call critical_region_exit(1), and interrupts
would not properly be reenabled. In the new version of the code,
we use our own nesting_count value to track the intended state, so
now nested disable()s only call critical_region_enter() once, only
updating is_nested_critical_region once; and only the second enable()
call will call critical_region_exit, with the right value of i_n_c_r.
Finally, in port_sleep_until_interrupt, if !sd_enabled, we really do
need to __disable_irq, rather than using the common_hal_mcu routines;
the reason why is documented in a comment.
This commit adds time.ticks_ms/us support using RTC1 as the timebase. It
also adds the time.ticks_add/diff helper functions. This feature can be
enabled using MICROPY_PY_TIME_TICKS. If disabled the system uses the
legacy sleep methods and does not have any ticks functions.
In addition support for MICROPY_EVENT_POLL_HOOK was added to the
time.sleep_ms(x) function, making this function more power efficient and
allows support for select.poll/asyncio. To support this, the RTC's CCR0
was used to schedule a ~1msec event to wakeup the CPU.
Some important notes about the RTC timebase:
- Since the granularity of RTC1's ticks are approx 30usec, time.ticks_us is
not perfect, does not have 1us resolution, but is otherwise quite usable.
For tighter measurments the ticker's 1MHz counter should be used.
- time.ticks_ms(x) should *not* be called in an IRQ with higher prio than
the RTC overflow irq (3). If so it introduces a race condition and
possibly leads to wrong tick calculations.
See #6171 and #6202.
new file: ports/nrf/boards/raytac_mdbt50q-db-40/bootloader/6.0.0/pca10056_bootloader_6.0.0_s140.zip
new file: ports/nrf/boards/raytac_mdbt50q-db-40/mpconfigboard.h
new file: ports/nrf/boards/raytac_mdbt50q-db-40/mpconfigboard.mk
new file: ports/nrf/boards/raytac_mdbt50q-db-40/pins.c
When compiling for debug (-O0) the .text segment cannot fit the flash
region when MICROPY_ROM_TEXT_COMPRESSION=1, because the compiler does not
optimise away the large if-else chain used to select the correct compressed
string.
This commit enforces MICROPY_ROM_TEXT_COMPRESSION=0 when compiling for
debug (DEBUG=1).
The storage space of the advertisement name is not declared static, leading
to a random advertisement name. This commit fixes the issue by declaring
it static.
The Bluetooth link gets disconnected when connecting from a PC after 30-40
seconds. This commit adds handling of the data length update request. The
data length parameter pointer is set to NULL in the reply, letting the
SoftDevice automatically set values and use them in the data length update
procedure.
mp_keyboard_interrupt() triggers a compiler error because the function is
implicitly declared. This commit adds "py/runtime.h" to the includes.
Fixes issue #5732.
Testing performed: That a card is successfully mounted on Pygamer with
the built in SD card slot
This module is enabled for most FULL_BUILD boards, but is disabled for
samd21 ("M0"), litex, and pca10100 for various reasons.
* Fix flash writes that don't end on a sector boundary. Fixes#2944
* Fix enum incompatibility with IDF.
* Fix printf output so it goes out debug UART.
* Increase stack size to 8k.
* Fix sleep of less than a tick so it doesn't crash.
This pulls all common functionality into `shared-bindings` and keeps
platform-specific code inside `nrf`. Additionally, this performs most
validation in the `shared-bindings` site.
The only validation that occurs inside platform-specific `common-hal`
code is related to timeout limits that are platform-specific.
Additionally, all documentation is now inside the `shared-bindings`
directory.
Signed-off-by: Sean Cross <sean@xobs.io>
The previous setting of `1` meant that the bluetooth system couldn't be
used when the watchdog timer was enabled.
Signed-off-by: Sean Cross <sean@xobs.io>
As part of the reset process, save the current tick count to an
uninitialized memory location. That way, the current tick value will be
preserved across reboots.
A reboot will cause us to lose a certain number of ticks, depending on
how long a reboot takes, however if reboots are infrequent then this
will not be a large amount of time lost.
Signed-off-by: Sean Cross <sean@xobs.io>
For `microcontroller.reset()`, don't manually call NVIC_SystemReset().
Instead, call the `port_reset()` in case the port wants to do any
cleanup.
Signed-off-by: Sean Cross <sean@xobs.io>
This finishes the rework of the exception handler, which is once
again stored inside the watchdog timer module.
This also implements a `watchdog_reset()` that is used to disable the
RAISE watchdog, if one is enabled.
Signed-off-by: Sean Cross <sean@xobs.io>
This adds an exception to be raised when the WatchDogTimer times out.
Note that this currently causes a HardFault, and it's not clear why it's
not behaving properly.
Signed-off-by: Sean Cross <sean@xobs.io>
This enables WDT support for Simmel. Other platforms cannot yet use
WDT because it overflows their flash storage.
Enable CIRCUITPY_WDT support for the nrf target.
Signed-off-by: Sean Cross <sean@xobs.io>
With the WDT changes, building Circuit Python results in the following error:
/opt/gcc-arm-none-eabi-9-2019-q4-major/bin/../lib/gcc/arm-none-eabi/9.2.1/../../../../arm-none-eabi/bin/ld: section .ARM.exidx LMA [00000000000621c8,00000000000621cf] overlaps section .data LMA [00000000000621c8,0000000000062383]
This is because unwinding data is getting generated, but has nowhere to go.
Re-enable this data in the linker script so it is saved.
Signed-off-by: Sean Cross <sean@xobs.io>
If interrupts are disabled, then calling sd_* functions will hardfault.
Instead, assume that it's safe to write if interrupts are disabled.
Signed-off-by: Sean Cross <sean@xobs.io>
ARM recommends issuing a DSB instruction propr to issuing WFI, as it is
required on many parts suchas Cortex-M7. This is effectively a no-op on
the Cortex-M4 used in most NRF parts, however it ensures that we won't
be surprised when new parts come out.
See
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dai0321a/BIHICBGB.html
for more information.
Signed-off-by: Sean Cross <sean@xobs.io>
In order to ensure we don't have any outstanding requests, disable
interrupts prior to issuing `WFI`.
As part of this process, check to see if there are any pending USB
requests, and only execute the `WFI` if there is no pending data.
This fixes#2855 on NRF.
Signed-off-by: Sean Cross <sean@xobs.io>
On NRF, the `rtc_reset()` function is never called. As a result,
calls to `time.time()` return a cryptic error>
```
>>> import time
>>> time.time()
'' object has no attribute 'datetime'
>>>
```
This is because `MP_STATE_VM(rtc_time_source)` is not initialized
due to `rtc_reset()` never being called.
If `CIRCUITPY_RTC` is enabled, call `rtc_reset()` as part of the
`reset_port()` call. This ensures that `time.time()` works as expected.
Signed-off-by: Sean Cross <sean@xobs.io>
The timeout value is calculated by the common-hal layer now, so we don't
need to be quite so clever about calculating it here.
Signed-off-by: Sean Cross <sean@xobs.io>
Add a field to allow specifying a timeout when initiating advertising.
As part of this, add a new property to determine if the device is still
advertising.
Additionally, have the `anonymous` property require a timeout, and set
the timeout to the maximum possible value if no timeout is specified.
Signed-off-by: Sean Cross <sean@xobs.io>
Add a new parameter to the `start_advertising()` function to enable
anonymous advertising. This forces a call to `sd_ble_gap_privacy_set()`
with `privacy_mode` set to `BLE_GAP_PRIVACY_MODE_DEVICE_PRIVACY` and
`private_addr_type` set to
`BLE_GAP_ADDR_TYPE_RANDOM_PRIVATE_RESOLVABLE`.
With this, addresses will cycle at a predefined rate (currently once
every 15 minutes).
Signed-off-by: Sean Cross <sean@xobs.io>
Store the RTC value in the .uninitialized section, but make sure to
flank it with some known values. That way we can determine if the RTC
value has been initialized, or if it's random uninitialized garbage.
As part of this, add a `common_hal_rtc_init()` routine to determine if
the value is correct, or reset it to 0 if it is not valid.
Signed-off-by: Sean Cross <sean@xobs.io>
Circuit Python supports saving a single word of data across reboots.
Previously, this data was placed immediately following the .bss.
However, this appeared to not work, as Circuit Python zeroes out the
heap when it starts up, and the heap begins immediately after the .bss.
Switch to using the new .uninitialized section in order to store this
word across resets.
Signed-off-by: Sean Cross <sean@xobs.io>
Previously, it was placed following .bss. However, now that there is a
new section after .bss, the heap must be moved forward.
Signed-off-by: Sean Cross <sean@xobs.io>
This section immediately follows the .bss section, and is designed to
contain uninitialized variables that should persist across reboots.
The section is placed directly after .bss, under the theory that the
size of Circuit Python's .bss + .data is bigger than the bootloader's
.bss + .data, so there is less likely to be a conflict.
Signed-off-by: Sean Cross <sean@xobs.io>
Commit 6cea369b89 updated the TinyUSB
submodule to a version based on nrfx v2.0.0. This commit updates the nrf
port to work with the latest TinyUSB and nrfx v2.0.0.
Because it can confuse older versions of gcc. Instead use the correct
instruction for Thumb vs Thumb-2 (sub vs subs) so the assembler emits the
2-byte instruction.
Related to commit 1aa9ff9141.
SPIM3 is faster than all other SPI blocks, and is capable of generating
a 32 MHz clock. However, it cannot be used at the same time as the BLE
radio without dedicating an additional 8 kB of RAM to it.
Therefore, some boards may want to disable this. Support pre-defining
NRFX_SPIM3_ENABLED on the command line to disable it on some bords.
Signed-off-by: Sean Cross <sean@xobs.io>
Simmel had USB HID disabled in order to save space. However, the board
configuration did not set USB_DEVICES, causing it to inherit the default
device bouquet of MSC,CDC,MIDI,HID. This in turn caused HID to be included
in the USB Configuration Descriptor.
For some reason, this was not a problem in an earlier version of tinyusb or
circuitpython. However, in the most recent version this has rightfully
caused asserts to appear during configuration.
Re-enable USB_HID for now, as it doesn't add too much in terms of space.
We may disable it again later on if we become pressed for space.
Signed-off-by: Sean Cross <sean@xobs.io>
This was not added as part of the initial v7.0.1 patchset because
.hex files are in this project's .gitignore, and so git ignored it.
Signed-off-by: Sean Cross <sean@xobs.io>
Note: the uncrustify configuration is explicitly set to 'add' instead of
'force' in order not to alter the comments which use extra spaces after //
as a means of indenting text for clarity.
Previously, we were using v6.1.1 which worked, but was unsupported.
v7.0.1 is the first version of s140 that supports the nRF52833.
Signed-off-by: Sean Cross <sean@xobs.io>
Allow for setting various softradio memory settings as part of a
board in order to support lower-memory configurations. If a
parameter is unspecified then the previously-defined value is used.
Signed-off-by: Sean Cross <sean@xobs.io>
Conditionally set variables such as the softdevice RAM size, bootloader
size, and the spi m3 buffer size. This allows ports to adjust these
values to suit their needs.
Signed-off-by: Sean Cross <sean@xobs.io>
The BLE Config area needs to be subtracted from the size of the firmware.
THis is because the firmware is counted by walking backwards from the end
of memory, and the BLE config area is placed lower in memory than the
firmware. Subtracting the BLE config size ensures the internal flash
filesystem doesn't try to use the firmware as storage.
Signed-off-by: Sean Cross <sean@xobs.io>
This adds preliminary support for the nRF52833, which is a variant of
the nRF52840 with half the RAM, half the flash, and fewer peripherals.
Signed-off-by: Sean Cross <sean@xobs.io>
Ports can set CIRCUITPY_RGBMATRIX and CIRCUITPY_FRAMEBUFFERIO to 0
in their .mk file in order to prevent these from being built. This
is necessary for resource-constrained devices.
Signed-off-by: Sean Cross <sean@xobs.io>
This gets all the purely internal references. Some uses of
protomatter/Protomatter/PROTOMATTER remain, as they are references
to symbols in the Protomatter C library itself.
I originally believed that there would be a wrapper library around it,
like with _pixelbuf; but this proves not to be the case, as there's
too little for the library to do.
Now that error string compression is supported it's more important to have
consistent error string formatting (eg all lowercase English words,
consistent contractions). This commit cleans up some of the strings to
make them more consistent.
This function is not used by the core but having it as part of the build
allows it to be used by user C modules, or board extensions. The linker
won't include it in the final firmware if it remains unused.
This string is recognised by uncrustify, to disable formatting in the
region marked by these comments. This is necessary in the qstrdef*.h files
to prevent modification of the strings within the Q(...). In other places
it is used to prevent excessive reformatting that would make the code less
readable.
The "random" module no longer uses the hardware RNG (the extmod version of
this module has a pseudo-random number generator), so the config option
MICROPY_PY_RANDOM_HW_RNG is no longer meaningful. This commit replaces it
with MICROPY_HW_ENABLE_RNG, which controls whether the hardware RNG is
included in the build.
This provides a more consistent C-level API to raise exceptions, ie moving
away from nlr_raise towards mp_raise_XXX. It also reduces code size by a
small amount on some ports.
If the exception doesn't need printf-style formatting then calling
mp_raise_msg is more efficient. Also shorten exception messages to match
style in core and other ports.
The backlight enable is active low on our board so the driver doesn't like it.
Toggling to GPIO manually for now.
As fixed the improper SPI bus pins definitions and it works!
Signed-off-by: Michael Welling <mwelling@ieee.org>
PacketBuffer facilitates packet oriented BLE protocols such as BLE
MIDI and the Apple Media Service.
This also adds PHY, MTU and connection event extension negotiation
to speed up data transfer when possible.
disable only turns off ENABLE but doesn't set the init tracking that
nrfx uses. uninit hangs if ENABLE is off and is called because it
waits forever for TX to stop.
For the 3 ports that already make use of this feature (stm32, nrf and
teensy) this doesn't make any difference, it just allows to disable it from
now on.
For other ports that use pyexec, this decreases code size because the debug
printing code is dead (it can't be enabled) but the compiler can't deduce
that, so code is still emitted.
.. a requirement that oofatfs needs to be taught to respect.
This problem can be demonstrated with the following snippet, except
that the related file ("test.bin") must also be contiguous on the
filesystem. You can ensure this by reformatting your device's filesystem
before testing, then copying any single file bigger than 4kB to test.bin.
f = open("test.bin", "rb")
f.seek(2048)
b = bytearray(2048)
v = memoryview(b)
f.readinto(v[909:])
Closes: #2332
The "spacing" of "buffer structure" is confusing, use the "channel count"
instead.
Testing performed on nrf52840 feather:
Play stereo and mono, 8- and 16-bit, 8kHz RawSamples representing 333.33Hz
square waves.
Use both mono and stereo PWMAudioOut instances.
Scope the RC-filtered signal and use the scope's frequency
measurement function, verify the frequency is 333 or 334Hz in all tested
cases.
In the "stereo output" cases, verify both the L and R channels. Verify
the output amplitude is the same in both channels.
In the "stereo output" cases, run a second test where the L channel's
amplitude is attenuated 50%. Verify the output amplitude is correct
in each channel.
The sample width register was never set, so all samples were played
as though they were 16 bit.
After this change, 8-bit samples no longer produce audio on the MAX 98357A
BOB, because only 16-, 24-, and 32-bit samples are supported by the
hardware. This will be addressed by a future change to pad samples to
16 bits; see #2323 and the 98357A datasheet page 6.
The meaning of the "single channel" parameter is not well-documented,
but in fact it seems that "true" must be passed or else the returned
channel_count is always 1. This caused stereo samples to be played
incorrectly.
This caused two problems when playing unsigned samples:
* When an even number of samples were present, it "worked" but only
every other sample was copied into the output, changing the waveform
* When an odd number of samples were present, the copy continued beyond
the end of the buffers and caused a hard fault
If we put no samples into the buffer, then there is no last
sample to fill out hold_value with. (and, in fact, the expression such
as *(uint32_t*)(buffer-4) is outside an allocated region)
Detect this condition, and leave the prior value in place.
This improves clicks heard when pausing and resuming a waveform.
This code is shared by most parts, except where not all the #ifdefs
inside the tick function were present in all ports. This mostly would
have broken gamepad tick support on non-samd ports.
The "ms32" and "ms64" variants of the tick functions are introduced
because there is no 64-bit atomic read. Disabling interrupts avoids
a low probability bug where milliseconds could be off by ~49.5 days
once every ~49.5 days (2^32 ms).
Avoiding disabling interrupts when only the low 32 bits are needed is a minor
optimization.
Testing performed: on metro m4 express, USB still works and
time.monotonic_ns() still counts up
This commit removes the Makefile-level MICROPY_FATFS config and moves the
MICROPY_VFS_FAT config to the Makefile level to replace it. It also moves
the include of the oofatfs source files in the build from each port to a
central place in extmod/extmod.mk.
For a port to enabled VFS FAT support it should now set MICROPY_VFS_FAT=1
at the level of the Makefile. This will include the relevant oofatfs files
in the build and set MICROPY_VFS_FAT=1 at the C (preprocessor) level.
This allows the board to disable the onboard speaker until explicitly
enabled in user code.
Testing performed on a CPB:
* Touching the AUDIO pin with a fingertip no longer generates noise/buzz
* Generating a waveform with `simpleio.tone` produces no sound by default
* When the board.SPEAKER_ENABLE is configured as a digital output and
set True, `simpleio.tone` does produce sound
Note that while guides should include information about SPEAKER_ENABLE, it's
possible that some users who omitted it could view this as a breaking change.
They can fix it by simply adding code similar to
speaker_enable = digitalio.DigitalInOut(board.SPEAKER_ENABLE)
speaker_enable.direction = digitalio.Direction.OUTPUT
speaker_enable.value = True
before using the internal speaker.
Closes: #2258
This commit implements automatic module weak links for all built-in
modules, by searching for "ufoo" in the built-in module list if "foo"
cannot be found. This means that all modules named "ufoo" are always
available as "foo". Also, a port can no longer add any other weak links,
which makes strict the definition of a weak link.
It saves some code size (about 100-200 bytes) on ports that previously had
lots of weak links.
Some changes from the previous behaviour:
- It doesn't intern the non-u module names (eg "foo" is not interned),
which saves code size, but will mean that "import foo" creates a new qstr
(namely "foo") in RAM (unless the importing module is frozen).
- help('modules') no longer lists non-u module names, only the u-variants;
this reduces duplication in the help listing.
Weak links are effectively the same as having a set of symbolic links on
the filesystem that is searched last. So an "import foo" will search
built-in modules first, then all paths in sys.path, then weak links last,
importing "ufoo" if it exists. Thus a file called "foo.py" somewhere in
sys.path will still have precedence over the weak link of "foo" to "ufoo".
See issues: #1740, #4449, #5229, #5241.
This PR refines the _bleio API. It was originally motivated by
the addition of a new CircuitPython service that enables reading
and modifying files on the device. Moving the BLE lifecycle outside
of the VM motivated a number of changes to remove heap allocations
in some APIs.
It also motivated unifying connection initiation to the Adapter class
rather than the Central and Peripheral classes which have been removed.
Adapter now handles the GAP portion of BLE including advertising, which
has moved but is largely unchanged, and scanning, which has been enhanced
to return an iterator of filtered results.
Once a connection is created (either by us (aka Central) or a remote
device (aka Peripheral)) it is represented by a new Connection class.
This class knows the current connection state and can discover and
instantiate remote Services along with their Characteristics and
Descriptors.
Relates to #586
.. otherwise, when an AudioPWMOut object was deinitted without being
explicitly stop()ped, it would use up a slot in active_audio[]; the
5th iteration would create a non-working audio object which would just
buzz instead of playing the right thing.
Closes: #2203
This patch add basic building blocks for nrf9P60.
It also includes a secure bootloader which forwards all
possible peripherals that are user selectable to become
non-secure. After configuring Flash, RAM and peripherals
the secure bootloader will jump to the non-secure domain
where MicroPython is placed.
The minimum size of a secure boot has to be a flash
block of 32Kb, hence why the linker scripts are
offsetting the main application this much.
The RAM offset is set to 128K, to allow for later
integration of Nordic Semiconductor's BSD socket
library which reserves the range 0x20010000 - 0x2001FFFF.
Add support for pca10059 with REPL over tinyusb USB CDC.
The board also includes a board specific module that will
recover UICR->REGOUT0 in case this has been erased.
This initial support does not preserve any existing bootloader
on the pca10090 in case this was present, and expects to use all
available flash on the device.
Add nrf-port finyusb driver files. USB CDC can be activated
by board configuration files using the MICROPY_HW_USB_CDC.
Updating BLE driver, Makefile, nrfx-glue and main.c to plug
in the tinyusb stack.
The the nrfx driver is aware of chip specific registers, while
the raw HAL abstraction is not. This driver enables use of NVMC
in non-secure domain for nrf9160.
This patch moves the check for MICROPY_PY_MACHINE_TEMP to come
before the inclusion of nrf_temp.h. The nrf_temp.h depends on
the NRF_TEMP_Type which might not be defined for all nRF devices.
@ladyada says:
"having this be adjustable (reference) would be ideal cause you can get
absolute voltages but for now, VCC/4 + 4x matches every other chip :)"
... and indeed doing it this way happens to give a much more steady
reading when using a VCC-referenced resistance, and so many of the simple
things you'd wire up are actually VCC-referenced anyway.
.. based on some tasks I found that caused stuttering:
# Test SD and printing
while True: os.listdir('.')
# Test bulk I/O
while True: len(open('somefile.wav', 'rb').read())
Each of these tasks *WAS* worse and I am improving them in a separate
PR by adding RUN_BACKGROUND_TASKS to them.
This enables the highest level of debug symbols, and all optimizations
except lto that do NOT interfere with debugging, in the view of the gcc
maintainers.
Testing performed: I used a Particle Xenon with a HDA1334 I2S DAC.
I played a variety of mono 16-bit samples at 11025 and 22050Hz nominal
bit rates. With this setup, all the 11025Hz samples sound good.
I tested play, pause, and loop functionality.
During some runs with 22050Hz samples, there were glitches. However,
these may have only occurred during runs where I had set breakpoints
and watchpoints in gdb.
I also tested with a MAX98357A I2S amplifier. On this device, everything
sounded "scratchy". I was powering it from 5V and the 5V rail seemed
steady, so I don't have an explanation for this. However, I haven't
tried it with a SAMD board.
It was previously not taking into account that the list of pins was sparse,
so using the wrong index. The boards/X/pins.csv was generating the wrong
data for machine.Pin.board.
As part of this fix rename the variables to make it more clear what the
list contains (only board pins).
mp_compile no longer takes an emit_opt argument, rather this setting is now
provided by the global default_emit_opt variable.
Now, when -X emit=native is passed as a command-line option, the emitter
will be set for all compiled modules (included imports), not just the
top-level script.
In the future there could be a way to also set this variable from a script.
Fixes issue #4267.
