thorstenhater / arbor-gui Goto Github PK

1. Some kind of syntax highlighting
2. Allow to build from templates
3. Type hinting

In particular on the RHS of the viewport, the handy Segment and Branch indicator disappears well before the cursor approaches the border of the viewport. E.g.:

Good:

Still would like to see Segment and Branch ID:

Sporadically the attached ASC file throws the following error, but usually it loads fine:

[2021-05-18 18:12:04.928] [debug] Reading ACC from /home/robin/stellate.asc
terminate called after throwing an instance of 'arborio::cableio_parse_error'
  what():  Unexpected term Unexpected '(' at the end of input. at :5:1
Aborted

stellate.zip

Make standard three light setup: Key, fill, back.

Cf. #45 - it would be nice if it would be possible to set the resolution of the snapshot.

Adding a locset with expression (terminal) draws triangles are every distal point of every segment.

Frustrum caps look weird due to lighting/normals. Add proper normals on the proper vertices.

• Github CI generated
  • https://github.com/marketplace/actions/build-appimage
• General AppImage packaging
  • https://github.com/AppImage/AppImageKit/

Right now rotating requires a pointer input with horizontal scrolling, i.e. a mousepad. For those without laptops, a UI control for this would be nice (and/or CTRL+vert scroll or sth). Panning is not possible, would also be nice (click+drag).

• need some differentiations from the actual geometry. Perhaps a black border?
• currently, a single skew triangle is drawn. Which is not very clear to read.

E.g. color scale for potential (per segment).

Using undefined names or unavailable segments in the regions or marker definitions lead to exceptions, e.g.,

terminate called after throwing an instance of 'arb::unbound_name'
  what():  no definition for 'proximal-dend-locs'

or

terminate called after throwing an instance of 'arb::no_such_segment'
  what():  segment 1545 out of bounds

These exceptions should be catched.

Given a morphology and a arb::cv_policy we should be able to show the CVs as a 3D rendering.

We can build everything in CI.

Putting the malformed locset (uniform () 0 9 0) in arbor-gui crashes thingify.

Arbor at 1ec287b54c6d1d42448d3824ea4c8b0776485a93.

[2021-05-12 13:28:23.138] [info] Making markers for locset Locset 1 '(uniform () 0 9 0)'

Thread 1 "arbor-gui" received signal SIGSEGV, Segmentation fault.
0x0000000000690306 in arb::ls::thingify_ (u=..., p=...) at /home/sschmitt/HBP/arbor-gui/3rd-party/arbor/arbor/morph/locset.cpp:442
442             auto cable = reg_cables[cable_idx];
(gdb) where
#0  0x0000000000690306 in arb::ls::thingify_ (u=..., p=...) at /home/sschmitt/HBP/arbor-gui/3rd-party/arbor/arbor/morph/locset.cpp:442
#1  0x000000000069c32b in arb::locset::wrap<arb::ls::uniform_>::thingify (this=0x151d210, m=...) at /home/sschmitt/HBP/arbor-gui/3rd-party/arbor/arbor/include/arbor/morph/locset.hpp:106
#2  0x0000000000470386 in arb::thingify (p=..., m=...) at /home/sschmitt/HBP/arbor-gui/3rd-party/arbor/arbor/include/arbor/morph/locset.hpp:63
#3  0x000000000046f77f in cell_builder::make_points (this=0x7fffffffd500, locset=...) at /home/sschmitt/HBP/arbor-gui/src/cell_builder.cpp:54
#4  0x0000000000482e30 in gui_state::event_visitor::operator() (this=0x7fffffffd388, c=...) at /home/sschmitt/HBP/arbor-gui/src/gui_state.cpp:1060
#5  0x0000000000487f41 in std::__invoke_impl<void, gui_state::update()::event_visitor, evt_upd_locdef<loc_def<arb::locset> >&>(std::__invoke_other, gui_state::event_visitor &&) (__f=...)
    at /home/sschmitt/spack/opt/spack/linux-ubuntu20.04-haswell/gcc-9.3.0/gcc-10.2.0-eily4eebiagzzlfboeezmfy3cei7zz3s/include/c++/10.2.0/bits/invoke.h:60
#6  0x000000000048715f in std::__invoke<gui_state::update()::event_visitor, evt_upd_locdef<loc_def<arb::locset> >&>(gui_state::event_visitor &&) (__fn=...)
    at /home/sschmitt/spack/opt/spack/linux-ubuntu20.04-haswell/gcc-9.3.0/gcc-10.2.0-eily4eebiagzzlfboeezmfy3cei7zz3s/include/c++/10.2.0/bits/invoke.h:95
#7  0x0000000000484ad9 in std::__detail::__variant::__gen_vtable_impl<std::__detail::__variant::_Multi_array<std::__detail::__variant::__deduce_visit_result<void> (*)(gui_state::update()::event_visitor&&, std::variant<evt_upd_cv, evt_add_mechanism, evt_del_mechanism, evt_add_ion, evt_del_ion, evt_add_probe, evt_del_probe, evt_add_stimulus, evt_del_stimulus, evt_add_detector, evt_del_detector, evt_add_locdef<loc_def<arb::region> >, evt_upd_locdef<loc_def<arb::region> >, evt_del_locdef<loc_def<arb::region> >, evt_add_locdef<loc_def<arb::locset> >, evt_upd_locdef<loc_def<arb::locset> >, evt_del_locdef<loc_def<arb::locset> > >&)>, std::integer_sequence<long unsigned int, 15> >::__visit_invoke(gui_state::event_visitor &&, std::variant<evt_upd_cv, evt_add_mechanism, evt_del_mechanism, evt_add_ion, evt_del_ion, evt_add_probe, evt_del_probe, evt_add_stimulus, evt_del_stimulus, evt_add_detector, evt_del_detector, evt_add_locdef<loc_def<arb::region> >, evt_upd_locdef<loc_def<arb::region> >, evt_del_locdef<loc_def<arb::region> >, evt_add_locdef<loc_def<arb::locset> >, evt_upd_locdef<loc_def<arb::locset> >, evt_del_locdef<loc_def<arb::locset> > > &) (__visitor=..., __vars#0=...) at /home/sschmitt/spack/opt/spack/linux-ubuntu20.04-haswell/gcc-9.3.0/gcc-10.2.0-eily4eebiagzzlfboeezmfy3cei7zz3s/include/c++/10.2.0/variant:1012
#8  0x0000000000484bdc in std::__do_visit<std::__detail::__variant::__deduce_visit_result<void>, gui_state::update()::event_visitor, std::variant<evt_upd_cv, evt_add_mechanism, evt_del_mechanism, evt_add_ion, evt_del_ion, evt_add_probe, evt_del_probe, evt_add_stimulus, evt_del_stimulus, evt_add_detector, evt_del_detector, evt_add_locdef<loc_def<arb::region> >, evt_upd_locdef<loc_def<arb::region> >, evt_del_locdef<loc_def<arb::region> >, evt_add_locdef<loc_def<arb::locset> >, evt_upd_locdef<loc_def<arb::locset> >, evt_del_locdef<loc_def<arb::locset> > >&>(gui_state::event_visitor &&) (__visitor=...) at /home/sschmitt/spack/opt/spack/linux-ubuntu20.04-haswell/gcc-9.3.0/gcc-10.2.0-eily4eebiagzzlfboeezmfy3cei7zz3s/include/c++/10.2.0/variant:1694
#9  0x0000000000484c37 in std::visit<gui_state::update()::event_visitor, std::variant<evt_upd_cv, evt_add_mechanism, evt_del_mechanism, evt_add_ion, evt_del_ion, evt_add_probe, evt_del_probe, evt_add_stimulus, evt_del_stimulus, evt_add_detector, evt_del_detector, evt_add_locdef<loc_def<arb::region> >, evt_upd_locdef<loc_def<arb::region> >, evt_del_locdef<loc_def<arb::region> >, evt_add_locdef<loc_def<arb::locset> >, evt_upd_locdef<loc_def<arb::locset> >, evt_del_locdef<loc_def<arb::locset> > >&>(gui_state::event_visitor &&) (
    __visitor=...) at /home/sschmitt/spack/opt/spack/linux-ubuntu20.04-haswell/gcc-9.3.0/gcc-10.2.0-eily4eebiagzzlfboeezmfy3cei7zz3s/include/c++/10.2.0/variant:1710
#10 0x0000000000484cc4 in gui_state::update (this=0x7fffffffd400) at /home/sschmitt/HBP/arbor-gui/src/gui_state.cpp:1155
#11 0x000000000047d656 in gui_state::gui (this=0x7fffffffd400) at /home/sschmitt/HBP/arbor-gui/src/gui_state.cpp:746
#12 0x000000000040da04 in main () at /home/sschmitt/HBP/arbor-gui/src/main.cpp:24

The cell window needs an indicator for the current orientation.

The default flavor for parsing SWC files is currently Neuron, I think it should be Arbor instead.
The traces page is the default page that is selected when launching the gui. I think it makes more sense for it to be cell instead.

When defining regions to be colored, it would be good if the order in which the coloring is done can be chosen. Like e.g. layers in a graphics program. Some level of transparency would also be helpful to define/debug definitions of overlapping regions.

Region that overlap are not necessarily painted in the way one would expect. A soluation involves formalizing what we expect, and then the technical challenge of painting to any one (part of a) region just once. Or alpha.

A small coordinate system and a scale (or even better a ruler function) would be nice to have.

It will help fix arbor-sim/arbor#1229

Currently the alignment of indented items is off by a few pixels, since the indentation is given in pixels. There should be a way to query ImGui for the proper values.

Along with arbor 0.6, we will release a first official version of the GUI.

Things to do until then:

• binary artificats
• Simulation handling
• Pull up to arbor 0.6
• Clean up ACC handling
  • CV policy SerDe
• Migrate to arbor-sim org
• Choose a better style
• Improve UX of viewport.
• Improve text editing UX.
• Add CLI options
  • Logging level.

No matter what my position in the scene is, snap to center always makes the cell disappear/somewhere far outside of viewport.

When I open the Probes section the following error occurs:

terminate called after throwing an instance of 'arb::no_such_mechanism'
  what():  no mechanism BBP in catalogue
Aborted

• If git tag starting with v is pushed to master, then create a Github Release with AppImage+Bundle from it.
• Have nightly builds from master (if master changed), generate a nightly Github release. See https://stackoverflow.com/questions/63014786/how-to-schedule-a-github-actions-nightly-build-but-run-it-only-when-there-where

Currently, we parse each region standalone, thus there can be no cross-references, eg a := (tag 1) b := (region "a").
Fix that by adding a temporary label-dict.

The following SWC morphology, loaded using Arbor flavor:

# id,  tag,      x,      y,      z,      r,    parent
    1     1     0.0     0.0     0.0     2.0        -1
    2     1    40.0     0.0     0.0     2.0         1
    3     3    40.0     0.0     0.0     2.0         2
    4     3   100.0     0.0     0.0     2.0         3

(segment 0) and (segment 1) overlap (segment 2). I would expect (segment 0) to be just the soma, (segment 2) to be just the dendrite and (segment 1) to be a zero-length segment.
I am using the Linux AppImage from the latest nightly release: https://github.com/thorstenhater/arbor-gui/actions/runs/1211712054

Please update documentation from

cmake .. -DCMAKE_CXX_COMPILER=/usr/local/opt/llvm/bin/clang++ \
         -DCMAKE_CCOMPILER=/usr/local/opt/llvm/bin/clang      \
         -DCMAKE_BUILD_TYPE=release

to

cmake .. -DCMAKE_CXX_COMPILER=/usr/local/opt/llvm/bin/clang++ \
         -DCMAKE_C_COMPILER=/usr/local/opt/llvm/bin/clang      \
         -DCMAKE_BUILD_TYPE=release

We have stb_image already, so just read the texture and store it as png.

Might be related to the platform (Linux): when I hold down CTRL and drag to pan, the viewport moves diagonally when the cursor is dragged at right angles, i.e. not in the direction of the cursor. Because this vector always seems to arc rightward, the cell tends to disappear on the LHS of the viewport.

Is it possible that the cell is anchored at the cursor position and dragged in the direction of the cursor, rather than/in addition to scroll? I suppose that was already the intent at [1], but there I can't see what might explain what I see.

[1] https://github.com/thorstenhater/arbor-gui/blob/master/src/window.cpp#L35

The executable needs an icon in the taskbars/docks and titlebars of this world.

• Titlebar (non-OSX only, not supported by OSX)
• Dock/executable icon on OSX
• Icon on Linux

glReadPixel is very slow, sometimes exceeding frame budgets.

Here's an idea:

• Shuffle the pick render to the front
• Define a PBO
• Initiate PBO transfer after pick render
• Finalise PBO transfer right before the end of the render step
• Or right before we read the value (that should be even later in the pipeline)

Currently colors are pulled from a pool of 12 pre-defined choices, which is not really scalable.
Should be reasonably doable via converting to HSV and stepping H by the GoldenRation.

To ease debugging in case of problems it would be helpful if the version, e.g., the commit hash plus dirty flag, could be retrieved from within the GUI. E.g. in the main window title or some "about" box. Command line usage via arbor-gui --version would also be good to have.

It would be nice if user defined names can be used in regions and locations, e.g. (from the examples),

# Regions:

labels['soma'] = '(tag 1)'
labels['axon'] = '(tag 2)'
labels['dend'] = '(tag 3)'
labels['last'] = '(tag 4)'

labels['all'] = '(all)'

labels['gt_1.5'] = '(radius-ge (region "all") 1.5)'
labels['custom'] = '(join (region "last") (region "gt_1.5"))'

# Locsets:

labels['root']     = '(root)'
labels['terminal'] = '(terminal)'
labels['custom_terminal'] = '(restrict (locset "terminal") (region "custom"))'
labels['axon_terminal'] = '(restrict (locset "terminal") (region "axon"))'

where all, axon, custom, etc. are used.

It would be handy to have an export/save-to-file feature available to avoid having to use screenshots. If possible, the resolution of the saved image could be different from the on screen render.

Using AppImage downloaded from here: https://github.com/thorstenhater/arbor-gui/actions/runs/1146342777
App crashes when a new region is defined as an empty s-expression().

arb::pwlin can return an arb::msegment{id, p', q'} describing a subsection of the 'original' segment (id, p, q) such that
p' = q + u(q - p) and q' = p + v(q - p) for u, v: [0, 1].

At the moment, the renderer just inspects the id and returns the full segment regardless of any points.

Idea

1. Add a field f to all points in the renderer that is 0 on the proximal and 1 on the distal end. Ensure linear interpolation.
2. Pass a pair (u, v) to each frustrum when rendering based on the fractions above. Ensure constant.
3. In the fragment shader discard if not u < f < v.

Issues

• Need to compute (u, v) from (p, q, p', q')
• Quite wasteful; usually we will show full segments, but have added 4 floats to each point

See: NeuroML/NeuroML2#144

To reproduce

1. Install arbor-gui
2. Download a NeuroML file (I used this one, or similarly this one)
3. File > Morphology > Load
4. Under Location > Regions, start deleting all regions that are not soma_group, dendrite_group or axon_group

Behaviour: At first, soma_group will change in random garbled text (might change continuously). After a while a segfault will occur

Suggestion: Add a branch for each supported version of arbor.

Opening morph.nml from Arbor's python/examples dir gives a segfault. Arbor-Gui configures Arbor with NML support, so, there is a possible problem. This is not with the AppImage btw, a good old CMake build installed to ~/.local, so no potential library problems. Output is not very helpful (straight to segfault). Have nml files been tested?