Screen class not working on ev3dev stretch

[ddemidov]

I am getting the following error with the recently released stretch package:

>>> import ev3dev.ev3 as ev3
>>> s = ev3.Screen()
>>> s.clear()
>>> s.update()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3/dist-packages/ev3dev/core.py", line 3320, in update
    self.mmap[:] = self._img.tobytes("raw", "1;IR")
IndexError: mmap slice assignment is wrong size

It seems the same error prevents openrobertalab service from working:

$ journalctl -n 100
Oct 28 10:08:34 ev3dev openrobertalab[1992]: INFO:roberta:--- starting ---
Oct 28 10:08:34 ev3dev openrobertalab[1992]: INFO:roberta.lab:python path: /usr/bin:/usr/lib/python35.zip:/usr/lib/python3.5:/usr/lib/python3.5/plat-arm-linux-gnueabi:/usr/lib/python3.5/lib-
Oct 28 10:08:34 ev3dev openrobertalab[1992]: INFO:roberta.lab:status changed: disconnected
Oct 28 10:08:38 ev3dev openrobertalab[1992]: Traceback (most recent call last):
Oct 28 10:08:38 ev3dev openrobertalab[1992]:   File "/usr/bin/openrobertalab", line 41, in <module>
Oct 28 10:08:38 ev3dev openrobertalab[1992]:     main()
Oct 28 10:08:38 ev3dev openrobertalab[1992]:   File "/usr/bin/openrobertalab", line 36, in main
Oct 28 10:08:38 ev3dev openrobertalab[1992]:     service = Service('/org/openroberta/Lab1')
Oct 28 10:08:38 ev3dev openrobertalab[1992]:   File "/usr/lib/python3/dist-packages/roberta/lab.py", line 88, in __init__
Oct 28 10:08:38 ev3dev openrobertalab[1992]:     self.hal.clearDisplay()
Oct 28 10:08:38 ev3dev openrobertalab[1992]:   File "/usr/lib/python3/dist-packages/roberta/ev3.py", line 179, in clearDisplay
Oct 28 10:08:38 ev3dev openrobertalab[1992]:     self.lcd.update()
Oct 28 10:08:38 ev3dev openrobertalab[1992]:   File "/usr/lib/python3/dist-packages/ev3dev/core.py", line 3320, in update
Oct 28 10:08:38 ev3dev openrobertalab[1992]:     self.mmap[:] = self._img.tobytes("raw", "1;IR")
Oct 28 10:08:38 ev3dev openrobertalab[1992]: IndexError: mmap slice assignment is wrong size
Oct 28 10:08:38 ev3dev openrobertalab[1992]: INFO:roberta:--- done ---

This is on the latest snapshot (snapshot-ev3dev-stretch-ev3-generic-2017-10-24.img.xz). The stretch-compatible ev3dev version is there, since leds are working:

>>> import ev3dev.ev3 as ev3
>>> ev3.Leds.all_off()

[ddemidov]

@dlech, were there any changes in framebuffer on stretch? For reference, here is what we currently do:

ev3dev-lang-python/ev3dev/core.py

Lines 3093 to 3234 in 02fa602

	class FbMem(object):
	"""The framebuffer memory object.
	Made of:
	- the framebuffer file descriptor
	- the fix screen info struct
	- the var screen info struct
	- the mapped memory
	"""
	# ------------------------------------------------------------------
	# The code is adapted from
	# https://github.com/LinkCareServices/cairotft/blob/master/cairotft/linuxfb.py
	#
	# The original code came with the following license:
	# ------------------------------------------------------------------
	# Copyright (c) 2012 Kurichan
	#
	# This program is free software. It comes without any warranty, to
	# the extent permitted by applicable law. You can redistribute it
	# and/or modify it under the terms of the Do What The Fuck You Want
	# To Public License, Version 2, as published by Sam Hocevar. See
	# http://sam.zoy.org/wtfpl/COPYING for more details.
	# ------------------------------------------------------------------
	__slots__ = ('fid', 'fix_info', 'var_info', 'mmap')
	FBIOGET_VSCREENINFO = 0x4600
	FBIOGET_FSCREENINFO = 0x4602
	FB_VISUAL_MONO01 = 0
	FB_VISUAL_MONO10 = 1
	class FixScreenInfo(ctypes.Structure):
	"""The fb_fix_screeninfo from fb.h."""
	_fields_ = [
	('id_name', ctypes.c_char * 16),
	('smem_start', ctypes.c_ulong),
	('smem_len', ctypes.c_uint32),
	('type', ctypes.c_uint32),
	('type_aux', ctypes.c_uint32),
	('visual', ctypes.c_uint32),
	('xpanstep', ctypes.c_uint16),
	('ypanstep', ctypes.c_uint16),
	('ywrapstep', ctypes.c_uint16),
	('line_length', ctypes.c_uint32),
	('mmio_start', ctypes.c_ulong),
	('mmio_len', ctypes.c_uint32),
	('accel', ctypes.c_uint32),
	('reserved', ctypes.c_uint16 * 3),
	]
	class VarScreenInfo(ctypes.Structure):
	class FbBitField(ctypes.Structure):
	"""The fb_bitfield struct from fb.h."""
	_fields_ = [
	('offset', ctypes.c_uint32),
	('length', ctypes.c_uint32),
	('msb_right', ctypes.c_uint32),
	]
	"""The fb_var_screeninfo struct from fb.h."""
	_fields_ = [
	('xres', ctypes.c_uint32),
	('yres', ctypes.c_uint32),
	('xres_virtual', ctypes.c_uint32),
	('yres_virtual', ctypes.c_uint32),
	('xoffset', ctypes.c_uint32),
	('yoffset', ctypes.c_uint32),
	('bits_per_pixel', ctypes.c_uint32),
	('grayscale', ctypes.c_uint32),
	('red', FbBitField),
	('green', FbBitField),
	('blue', FbBitField),
	('transp', FbBitField),
	]
	def __init__(self, fbdev=None):
	"""Create the FbMem framebuffer memory object."""
	fid = FbMem._open_fbdev(fbdev)
	fix_info = FbMem._get_fix_info(fid)
	fbmmap = FbMem._map_fb_memory(fid, fix_info)
	self.fid = fid
	self.fix_info = fix_info
	self.var_info = FbMem._get_var_info(fid)
	self.mmap = fbmmap
	def __del__(self):
	"""Close the FbMem framebuffer memory object."""
	self.mmap.close()
	FbMem._close_fbdev(self.fid)
	@staticmethod
	def _open_fbdev(fbdev=None):
	"""Return the framebuffer file descriptor.
	Try to use the FRAMEBUFFER
	environment variable if fbdev is not given. Use '/dev/fb0' by
	default.
	"""
	dev = fbdev or os.getenv('FRAMEBUFFER', '/dev/fb0')
	fbfid = os.open(dev, os.O_RDWR)
	return fbfid
	@staticmethod
	def _close_fbdev(fbfid):
	"""Close the framebuffer file descriptor."""
	os.close(fbfid)
	@staticmethod
	def _get_fix_info(fbfid):
	"""Return the fix screen info from the framebuffer file descriptor."""
	fix_info = FbMem.FixScreenInfo()
	fcntl.ioctl(fbfid, FbMem.FBIOGET_FSCREENINFO, fix_info)
	return fix_info
	@staticmethod
	def _get_var_info(fbfid):
	"""Return the var screen info from the framebuffer file descriptor."""
	var_info = FbMem.VarScreenInfo()
	fcntl.ioctl(fbfid, FbMem.FBIOGET_VSCREENINFO, var_info)
	return var_info
	@staticmethod
	def _map_fb_memory(fbfid, fix_info):
	"""Map the framebuffer memory."""
	return mmap.mmap(
	fbfid,
	fix_info.smem_len,
	mmap.MAP_SHARED,
	mmap.PROT_READ | mmap.PROT_WRITE,
	offset=0
	)

[dlech]

There is now 2bpp support in the driver, so fix_info.smem_len is probably twice as big as it was previously.

Could you change

self.mmap[:] = self._img.tobytes("raw", "1;IR")

to

b = self._img.tobytes("raw", "1;IR")
self.mmap[:len(b)] = b

?

[ddemidov]

That works, although I had to invert the image (or rather stop inverting it, by using "1;R" instead of "1;IR").
The following patch works for me:

diff --git a/ev3dev/core.py b/ev3dev/core.py
index 0dd29ca..72d0f41 100644
--- a/ev3dev/core.py
+++ b/ev3dev/core.py
@@ -3317,7 +3317,8 @@ class Screen(FbMem):
         Nothing will be drawn on the screen until this function is called.
         """
         if self.var_info.bits_per_pixel == 1:
-            self.mmap[:] = self._img.tobytes("raw", "1;IR")
+            b = self._img.tobytes("raw", "1;R")
+            self.mmap[:len(b)] = b
         elif self.var_info.bits_per_pixel == 16:
             self.mmap[:] = self._img_to_rgb565_bytes()
         else:

I am bit surprised by the result though. tobytes("raw", "1;R") should give us raw 1-bit-per-pixel image. Why assigning the result to a 2-bit-per-pixel framebuffer works?

[dlech]

The fbdev device has two modes of operation, 1bpp and 2bpp. Be default, it still operates in 1bpp mode. The memory allocated has to be big enough for both modes, so the 1bpp mode only uses 1/2 of the allocated memory.

[ddemidov]

Ok, that makes sense. I'll create PRs for stretch-stable and develop then.

[dlech]

I'm wondering about the inversion though. I don't recall switching from FB_VISUAL_MONO10 to FB_VISUAL_MONO01 but maybe I did.

[ddemidov]

Do you need more info from me on this? The EV3 is next to me right now.

[dlech]

The driver still says FB_VISUAL_MONO10, so I would expect that the inversion is still needed.

https://github.com/ev3dev/lego-linux-drivers/blob/ev3dev-stretch/linux/fbdev/st7586fb.c#L483

[ddemidov]

Well, without the inversion s.clear() gives me black screen, and s.draw.rectangle((10,10,168,118)) gives me a white-on-black rectangle.

[dlech]

ev3dev-lang-python/ev3dev/display.py

Line 257 in 8e57eea

self._draw.rectangle(((0, 0), self.shape), fill="white")

says clear should fill with white, not black

[ddemidov]

I believe that is because we had to do inversion.