The Smell of Molten Projects in the Morning

Ed Nisley's Blog: Shop notes, electronics, firmware, machinery, 3D printing, laser cuttery, and curiosities. Contents: 100% human thinking, 0% AI slop.

Category: Software

General-purpose computers doing something specific

  • Quilting Hexagon Template Generator

    Quilting Hexagon Template Generator

    Mary took on the task of finishing a hexagonal quilt from pieced strips, only to discover she’ll need several more strips and the myriad triangles required to turn hexagons into strips. The as-built strips do not match any of the standard pattern sizes, which meant ordinary templates were unavailing. I offered to build a template matching the (average) as-built hexagons, plus a triangle template based on those dimensions.

    Wikipedia has useful summaries of hexagon and equilateral triangle geometry and equations.

    Quilters measure hexes based on their finished side length, so a “1 inch hex” has sides measuring 1 inch, with the seam allowance extending ¼ inch beyond the sides. It’s difficult to measure finished sides with sufficient accuracy, so we averaged the side-to-side distance across several hexes.

    Some thrashing around produced a quick-and-dirty check piece that matched (most of) the stack of un-sewn hexes:

    Quilting Hexagon Cutting Template
    Quilting Hexagon Cutting Template

    That one came from a knockoff of the circle template, after some cleanup & tweakage, but failed user testing for not withstanding the side force from the rotary cutter blade. The inside and outside dimensions were correct, however, so I could proceed with some confidence I understood the geometry.

    Both the pattern width (the side-to-side distance across the inside of the hex) and the seam allowance appearing in the Customizer appear in inches, because that’s how things get measured outside the Basement Laboratory & Fabrication Facility:

    FinishedWidthInch = 2.75;
    FinishedWidth = FinishedWidthInch * inch;
    
    SeamAllowanceInch = 0.25;
    SeamAllowance = SeamAllowanceInch * inch;

    You feed in one side-to-side measurement and all other hex dimensions get calculated from that number; quilters default to a ¼ inch seam allowance. Remember, standard quilt hexes are measured by their side length, so just buy some standard templates.

    This is one of the few times I’ve needed triangle graph paper:

    Hex Quilting Template - geometry doodles
    Hex Quilting Template – geometry doodles

    After I gave up trying to get it right on square-grid paper, of course.

    Solidifying those relations:

    Quilting Hex Template - build layout
    Quilting Hex Template – build layout

    Then math got real:

    Hex Quilting Templates - on strips
    Hex Quilting Templates – on strips

    Both templates have non-skid strips to keep the fabric in place while cutting:

    Hex Quilting Template - grip strips
    Hex Quilting Template – grip strips

    I should have embossed the size on each template, but this feels like a one-off project and YAGNI. Of course, that’s how I felt about the circle templates, so maybe next time I’ll get it right.

    As it turned out, Mary realized she needed a template for the two half-triangles at the end of each row:

    Quilting Hex Template - half-triangle
    Quilting Hex Template – half-triangle

    It’s half of the finished size of the equilateral triangle on the right, with seam allowance added all around. The test scrap of fabric on the left shows the stitching along the hypotenuse of the half-triangle, where it joins to the end-of-row hexagon. Ideally, you need two half-triangle templates, but Mary says it’s easier to cut the fabric from the back side than to keep track of two templates.

    The family portrait now has three members:

    Quilting Hex Template - family
    Quilting Hex Template – family

    The OpenSCAD source code as a GitHub Gist:

    // Quilting – Hexagon Templates
    // Ed Nisley KE4ZNU – July 2020
    // Reverse-engineered to repair a not-quite-standard hexagon quilt
    // Useful geometry:
    // https://en.wikipedia.org/wiki/Hexagon
    /* [Layout Options] */
    Layout = "Build"; // [Build, HexBuild, HexPlate, TriBuild, TriPlate, EndBuild, EndPlate]
    //——-
    //- Extrusion parameters must match reality!
    // Print with 2 shells
    /* [Hidden] */
    ThreadThick = 0.25;
    ThreadWidth = 0.40;
    HoleFinagle = 0.2;
    HoleFudge = 1.00;
    function HoleAdjust(Diameter) = HoleFudge*Diameter + HoleFinagle;
    Protrusion = 0.1; // make holes end cleanly
    function IntegerMultiple(Size,Unit) = Unit * ceil(Size / Unit);
    inch = 25.4;
    //——-
    // Dimensions
    /* [Layout Options] */
    FinishedWidthInch = 2.75;
    FinishedWidth = FinishedWidthInch * inch;
    SeamAllowanceInch = 0.25;
    SeamAllowance = SeamAllowanceInch * inch;
    TemplateThick = 3.0;
    TriKnob = true;
    EndKnob = false;
    /* [Hidden] */
    FinishedSideInch = FinishedWidthInch/sqrt(3);
    FinishedSide = FinishedSideInch * inch;
    echo(str("Finished side: ",FinishedSideInch," inch"));
    CutWidth = FinishedWidth + 2*SeamAllowance;
    CutSide = CutWidth/sqrt(3);
    echo(str("Cut side: ",CutSide / inch," inch"));
    // Make polygon-circles circumscribe the target widths
    TemplateID = FinishedWidth / cos(180/6);
    TemplateOD = CutWidth / cos(180/6);
    /* [Hidden] */
    TriRadius = FinishedSide/sqrt(3);
    TriPoints = [[TriRadius,0],
    [TriRadius*cos(120),TriRadius*sin(120)],
    [TriRadius*cos(240),TriRadius*sin(240)]
    ];
    echo(str("TriPoints: ",TriPoints));
    EndPoints = [[TriRadius,0],
    [TriRadius*cos(120),TriRadius*sin(120)],
    [TriRadius*cos(120),0]
    ];
    echo(str("EndPoints: ",EndPoints));
    TipCutRadius = 2*(TriRadius + SeamAllowance); // circumscribing radius of tip cutter
    TipPoints = [[TipCutRadius,0],
    [TipCutRadius*cos(120),TipCutRadius*sin(120)],
    [TipCutRadius*cos(240),TipCutRadius*sin(240)]
    ];
    HandleHeight = 1 * inch;
    HandleLength = (TemplateID + TemplateOD)/2;
    HandleThick = IntegerMultiple(3.0,ThreadWidth);
    HandleSides = 12*4;
    StringDia = 4.0;
    StringHeight = 0.6*HandleHeight;
    DentDepth = HandleThick/4;
    DentDia = 15 * DentDepth;
    DentSphereRadius = (pow(DentDepth,2) + pow(DentDia,2)/4)/(2*DentDepth);
    KnobOD = 15.0; // Triangle handle
    KnobHeight = 20.0;
    //——-
    module PolyCyl(Dia,Height,ForceSides=0) { // based on nophead's polyholes
    Sides = (ForceSides != 0) ? ForceSides : (ceil(Dia) + 2);
    FixDia = Dia / cos(180/Sides);
    cylinder(r=HoleAdjust(FixDia)/2,h=Height,$fn=Sides);
    }
    //——-
    // Hex template
    module HexPlate() {
    difference() {
    cylinder(r=TemplateOD/2,h=TemplateThick,$fn=6);
    translate([0,0,-Protrusion])
    cylinder(r=TemplateID/2,h=(TemplateThick + 2*Protrusion),$fn=6);
    }
    for (i=[1:6/2])
    rotate(i*60)
    translate([0,0,TemplateThick/2])
    cube([HandleLength,HandleThick,TemplateThick],center=true);
    }
    module HexHandle() {
    difference() {
    rotate([90,0,0])
    scale([1,HandleHeight/(TemplateOD/2),1])
    rotate(180/HandleSides)
    cylinder(d=HandleLength,h=HandleThick,center=true,$fn=HandleSides);
    translate([0,0,-HandleHeight])
    cube([2*TemplateOD,2*TemplateOD,2*HandleHeight],center=true);
    translate([0,HandleThick,StringHeight])
    rotate([90,090,0])
    rotate(180/8)
    PolyCyl(StringDia,2*HandleThick,8);
    for (j=[-1,1]) {
    translate([0,j*(DentSphereRadius + HandleThick/2 – DentDepth),StringHeight])
    rotate(180/48)
    sphere(r=DentSphereRadius,$fn=48);
    }
    }
    }
    module HexTemplate() {
    HexPlate();
    HexHandle();
    }
    //——-
    // Triangle template
    module TriPlate() {
    linear_extrude(height=TemplateThick)
    intersection() {
    offset(delta=SeamAllowance) // basic cutting outline
    polygon(points=TriPoints);
    rotate(180)
    polygon(points=TipPoints);
    }
    }
    module TriTemplate() {
    union() {
    if (TriKnob)
    cylinder(d=KnobOD,h=KnobHeight,$fn=HandleSides);
    TriPlate();
    }
    }
    //——-
    // End piece template
    module EndPlate() {
    linear_extrude(height=TemplateThick)
    intersection() {
    offset(delta=SeamAllowance) // basic cutting outline
    polygon(points=EndPoints);
    rotate(180)
    polygon(points=TipPoints);
    }
    }
    module EndTemplate() {
    union() {
    if (EndKnob)
    translate([0,(TriRadius/2)*sin(30),0])
    cylinder(d=KnobOD,h=KnobHeight,$fn=HandleSides);
    EndPlate();
    }
    }
    //——-
    // Build it!
    if (Layout == "HexPlate")
    HexPlate();
    if (Layout == "HexBuild")
    HexTemplate();
    if (Layout == "TriPlate")
    TriPlate();
    if (Layout == "TriBuild")
    TriTemplate();
    if (Layout == "EndPlate")
    EndPlate();
    if (Layout == "EndBuild")
    EndTemplate();
    if (Layout == "Build") {
    translate([1.5*TriRadius,-TriRadius,0])
    rotate(180/6)
    TriTemplate();
    translate([-1.5*TriRadius,-TriRadius,0])
    rotate(180/6)
    EndTemplate();
    translate([0,TemplateOD/2,0])
    HexTemplate();
    }

  • Closing the Dmesg Audit Firehose

    I’m not entirely clear what’s being audited in the Manjaro Linux boxes I’ve recently set up, nor what the difference between res=success and res=failed might mean for the x11vnc unit:

    [   98.632347] audit: type=1131 audit(1594859418.419:110): pid=1 uid=0 auid=4294967295 ses=4294967295 msg='unit=x11vnc comm="systemd" exe="/usr/lib/systemd/systemd" hostname=? addr=? terminal=? res=success'
    [   98.632348] audit: audit_lost=46 audit_rate_limit=0 audit_backlog_limit=64
    [   98.632349] audit: kauditd hold queue overflow
    [   98.649743] audit: type=1130 audit(1594859418.433:111): pid=1 uid=0 auid=4294967295 ses=4294967295 msg='unit=x11vnc comm="systemd" exe="/usr/lib/systemd/systemd" hostname=? addr=? terminal=? res=failed'
    [   98.649745] audit: audit_lost=47 audit_rate_limit=0 audit_backlog_limit=64
    [   98.649746] audit: kauditd hold queue overflow
    [  100.515527] audit: type=1101 audit(1594859420.299:112): pid=843 uid=1000 auid=1000 ses=2 msg='op=PAM:accounting grantors=pam_unix,pam_permit,pam_time acct="ed" exe="/usr/bin/sudo" hostname=? addr=? terminal=/dev/pts/0 res=success'

    That’s the better part of two seconds in the life of the box and, later on, the pace picks up. Casual searching suggests nobody else knows what’s going on, either, apart from the fact than that it obviously has something to do with systemd and, thus, is just the way things are these days.

    Add audit=off to the default kernel command-line parameters by editing /etc/default/grub thusly:

    GRUB_CMDLINE_LINUX_DEFAULT="quiet resume=UUID=whatever udev.log_priority=3 audit=off"

    Then update the bootloader:

    sudo grub-mkconfig -o /boot/grub/grub.cfg 

    Whereupon dmesg becomes all quiet and (mostly) meaningful to this civilian.

  • Copying All! The! Files! (Except Some)

    All our data files spin around on a nearly full 1 TB drive in a “file server”, a grandiosely overqualified and dirt-cheap off-lease Dell Optiplex desktop sitting in the basement. It’s been running headless and unattended for the last half-dozen years and is badly in need of replacement, so I must copy all its files to a newer, even more overqualified, and equally cheap off-lease Optiplex.

    Copying the files from the /mnt/music collection on the existing server to the identically named directory on the new server proceeds thusly:

    sudo mount -o ro fileserver.local:/mnt/music /mnt/nfs
    sudo rsync -ahu --progress --log-file=/tmp/music.log \
     --exclude="/lost+found" \
     --exclude=".Trash*" \
     --exclude=".dtrash*" \
     --delete \
     /mnt/nfs/ /mnt/music

    Mount the existing collection (from the old server) in read-only mode to avoid heartache subsequent to confusion. It could happen.

    The first time through, add a -n option for a dry run, then inspect the log file for surprises.

    The various --exclude options avoid copying trashed-but-not-yet-deleted files from the various trash directories maintained by various file handlers. In the process of sorting this out, I learned the DigiKam photo manager creates a .dtrash directory holding deleted files for each of its Album listings, appearing down near the bottom of the top-level album wherein you’ve quasi-deleted photos via “Move to Trash”.

    The --delete option removes files on the destination (new disk) if they’re not on the source (old disk). I started this migration earlier this year, before the world fell apart, and have moved / consolidated / renamed various directories & files in the interim, so deleting the previous copies from their old locations makes the destination match the source.

    So far, so good …

  • Bathroom Door Retainer: Bigger and Stronger

    Bathroom Door Retainer: Bigger and Stronger

    After three years, the retainer holding the front bathroom door open against winds blowing through the house on stormy days finally fractured, right at the top of the towel rack where you’d expect it:

    Bathroom Door Retainer - fractured
    Bathroom Door Retainer – fractured

    I was all set to add reinforcing pins and whatnot, then came to my senses and just made the whole thing a few millimeters larger:

    Bathroom Door Retainer - stronger
    Bathroom Door Retainer – stronger

    Customer feedback indicates white blends better with the background.

    I made a few minor tweaks to the original design, including slightly larger bumps to hold it against the towel bar that, regrettably, put corresponding gouges into the bar. Who knew they used such soft plastic back in the day?

    The OpenSCAD source code as a GitHub Gist:

    // Bathroom Door Retainer
    // Ed Nisley KE4ZNU – May 2017
    // 2020-07 beef up, particularly at top of bar
    Layout = "Show"; // [Show, Build]
    //——-
    //- Extrusion parameters must match reality!
    /* [Hidden] */
    ThreadThick = 0.20;
    ThreadWidth = 0.40;
    HoleWindage = 0.2;
    Protrusion = 0.1; // make holes end cleanly
    function IntegerMultiple(Size,Unit) = Unit * ceil(Size / Unit);
    //——-
    // Dimensions
    /* [Dimensions] */
    TowelBarSide = 20.5; // towel bar across flat side
    TowelBarAngle = 45; // rotation of top flat from horizontal
    BumpOD = 2.0; // retaining ball
    DoorOffset = 14.0; // from towel bar to door
    DoorThick = 37.0;
    WallThick = 8.0; // minimum wall thickness
    PlateThick = 4.0; // … slab
    RetainerDepth = 15.0; // thickness of retaining notch
    NumSides = 6*4;
    CornerRad = WallThick;
    BarClipOD = TowelBarSide*sqrt(2) + 2*WallThick;
    BarClipRad = BarClipOD/2;
    OAH = RetainerDepth + PlateThick;
    module LatchPlan() {
    union() {
    linear_extrude(height=OAH,convexity=4)
    difference() {
    union() {
    circle(d=BarClipOD,$fn=NumSides);
    hull()
    for (i=[0,1], j=[0,1])
    translate([i*(BarClipRad + DoorOffset + DoorThick + WallThick – CornerRad),j*(BarClipRad – CornerRad)])
    circle(r=CornerRad,$fn=4*4);
    }
    rotate(TowelBarAngle) // towel bar shape
    square(size=TowelBarSide,center=true);
    translate([0,-TowelBarSide/sqrt(2)]) // make access slot
    rotate(-TowelBarAngle)
    square(size=[2*TowelBarSide,TowelBarSide],center=false);
    }
    for (a=[0:180:360])
    rotate(a + TowelBarAngle)
    translate([TowelBarSide/2,0,OAH/2])
    rotate([90,0,45])
    sphere(d=BumpOD,$fn=4*3);
    }
    }
    module Latch() {
    difference() {
    LatchPlan();
    translate([BarClipRad + DoorOffset,-BarClipRad/2,-Protrusion])
    cube([DoorThick,BarClipOD,RetainerDepth + Protrusion],center=false);
    }
    }
    //——-
    // Build it!
    if (Layout == "Show") {
    Latch();
    }
    if (Layout == "Build") {
    translate([0,0,OAH])
    rotate([180,0,0])
    Latch();
    }

    Done!

  • Manjaro Linux vs. Dell Latitude E7250 Bluetooth

    Manjaro Linux vs. Dell Latitude E7250 Bluetooth

    Although the Dell Latitude E7250 allegedly had Bluetooth capability and the Blueman applet tried connecting to my Bluetooth headsets, the connection aways failed and nothing worked. There’s a WLAN module stuck in an M.2 socket inside the laptop providing both WiFi and Bluetooth:

    Dell E7250 - DW1560 card in place
    Dell E7250 – DW1560 card in place

    A bit of searching suggested the driver wasn’t loading properly, which became obvious after I knew where to look:

    dmesg | grep -i blue
    … snippage …
    [    5.678610] Bluetooth: hci0: BCM20702A1 (001.002.014) build 1572
    [    5.678851] bluetooth hci0: Direct firmware load for brcm/BCM20702A1-0a5c-216f.hcd failed with error -2
    [    5.678853] Bluetooth: hci0: BCM: Patch brcm/BCM20702A1-0a5c-216f.hcd not found
    [   10.854607] Bluetooth: RFCOMM TTY layer initialized
    [   10.854613] Bluetooth: RFCOMM socket layer initialized
    [   10.854619] Bluetooth: RFCOMM ver 1.11

    Without having the proper firmware / patch loaded, the module won’t work, even though the TTY / socket layers know it’s present, which explains why Blueman did everything except actually connect to the headsets.

    More searching suggested you must extract the firmware HEX file from the Windows driver. Feeding the Service Tag into the Dell support site, then feeding “Bluetooth” and “Windows 8.1, 64-bit” (preinstalled on the laptop) into the Drivers & Downloads tab gets you the relevant EXE file: Dell Wireless 1550/1560 Wi-Fi and Bluetooth Driver. It turns out to be a self-extracting ZIP file (in Windows, anyway), so unzip it all by yourself:

    unzip Network_Driver_5DFVH_WN32_6.30.223.262_A03.EXE

    This produces a blizzard of HEX files in the newly created Drivers/production/Windows8.1-x64 directory. Each firmware HEX file is keyed to the USB Product Code identifying the unique USB gadget, found with lsusb:

    lsusb
    … snippage …
    Bus 002 Device 003: ID 0a5c:216f Broadcom Corp. BCM20702A0 Bluetooth
    … snippage …

    The DW1560 apparently has a USB RAM interface, with the specific HEX file identified in the CopyList stanza of the INF file corresponding to that USB Product Code:

    grep -i -A 5  ramusb216f.copylist Drivers/production/Windows8.1-x64/bcbtums-win8x64-brcm.inf
    [RAMUSB216F.CopyList]
    bcbtums.sys
    btwampfl.sys
    BCM20702A1_001.002.014.1443.1572.hex
    … snippage …

    However, the Linux firmware loader needs a different file format with a different name, mashed together from the HEX file, USB Vendor, and USB Product codes:

    hex2hcd -o BCM20702A1-0a5c-216f.hcd BCM20702A1_001.002.014.1443.1572.hex

    The converted firmware file goes where the loader expected to find it:

    sudo cp BCM20702A1-0a5c-216f.hcd /lib/firmware/brcm/

    Whereupon next reboot sorted things out:

    dmesg | grep -i blue
    [    6.024838] Bluetooth: Core ver 2.22
    [    6.024868] Bluetooth: HCI device and connection manager initialized
    [    6.024872] Bluetooth: HCI socket layer initialized
    [    6.024874] Bluetooth: L2CAP socket layer initialized
    [    6.024881] Bluetooth: SCO socket layer initialized
    [    6.100796] Bluetooth: BNEP (Ethernet Emulation) ver 1.3
    [    6.100800] Bluetooth: BNEP filters: protocol multicast
    [    6.100804] Bluetooth: BNEP socket layer initialized
    [    6.157114] Bluetooth: hci0: BCM: chip id 63
    [    6.158125] Bluetooth: hci0: BCM: features 0x07
    [    6.176119] Bluetooth: hci0: BCM20702A
    [    6.177114] Bluetooth: hci0: BCM20702A1 (001.002.014) build 0000
    [    7.031228] Bluetooth: hci0: BCM20702A1 (001.002.014) build 1572
    [    7.047177] Bluetooth: hci0: DW1560 Bluetooth 4.0 LE
    [   13.141854] Bluetooth: RFCOMM TTY layer initialized
    [   13.141865] Bluetooth: RFCOMM socket layer initialized
    [   13.141872] Bluetooth: RFCOMM ver 1.11

    The firmware may be in one of the myriad Bluetooth packages not installed by default, so perhaps identifying & installing the proper package would sidestep the hocus-pocus.

    Maybe next time?

    Now I can wear my Bose Hearphones in Zoom sessions with the E7250, because my Pixel 3a phone heats up almost to the gets-bendy level while thrashing its battery to death.

  • JPG Recovery From a Camera FAT Filesystem

    JPG Recovery From a Camera FAT Filesystem

    You can do it by hand, as I used to, or use recoverjpeg:

    dmesg | tail
    cd /tmp
    sudo dcfldd if=/dev/sde1 of=pix.bin bs=1M count=100
    recoverjpeg pix.bin 
    ristretto image00*
    

    Nothing prizewinning, but better than no picture at all:

    Garage Robin - recovered image
    Garage Robin – recovered image

    Note that you start by copying a reasonable chunk of the partition from the Memory Stick / (micro)SD Card first, to prevent a bad situation from getting worse.

    Now I can remember the easy way the next time around this block …

  • Solid Modeling: Support Puzzle

    Solid Modeling: Support Puzzle

    I’ve been putting this type of support structure inside screw holes & suchlike for years:

    Browning Hi-Power Magazine Block - solid model - Generic 1 - support detail
    Browning Hi-Power Magazine Block – solid model – Generic 1 – support detail

    It’s basically a group of small rectangles rotated around the hole’s axis and about one thread thickness shorter than the overhanging interior.

    I’ve found that incorporating exactly the right support structure eliminates Slic3r’s weird growths, eases removal, and generally works better all around.

    So doing this for the baseplate of the Glass Tile frame came naturally:

    Glass Tile Frame - octagonal support
    Glass Tile Frame – octagonal support

    This OpenSCAD snippet plunks one of those asterisks in each of four screw holes:

      if (Support)
        color("Yellow")
          for (i=[-1,1], j=[-1,1])
            translate([i*InsertOC.x/2,j*InsertOC.y/2,0])
              for (a=[0:45:135])
                  rotate(a)
                    translate([0,0,(Screw[LENGTH] - ThreadThick)/2])
                      cube([Screw[OD] - 2*ThreadWidth,2*ThreadWidth,Screw[LENGTH] - ThreadThick],center=true);

    The “cubes” overlap in the middle, with no completely coincident faces or common edges, so it’s 2-manifold. Slic3r, however, produces a weird time estimate whenever the model includes those structures:

    Slic3r - NaN time estimate
    Slic3r – NaN time estimate

    NaN stands for Not A Number and means something horrible has happened in the G-Code generation. Fortunately, the G-Code worked perfectly and produced the desired result, but I’m always uneasy when Something Seems Wrong.

    Messing around with the code produced a slightly different support structure:

    Glass Tile Frame - quad support
    Glass Tile Frame – quad support

    The one thread thick square on the bottom helps glue the structure to the platform and four ribs work just as well as eight in the octagonal hole:

      Fin = [Screw[OD]/2 - 1.5*ThreadWidth,2*ThreadWidth,ScrewRecess - ThreadThick];
      if (Inserts && SupportInserts)
        color("Yellow")
          for (i=[-1,1], j=[-1,1])
            translate([i*InsertOC.x/2,j*InsertOC.y/2,0]) {
              rotate(180/8)
                cylinder(d=6*ThreadWidth,h=ThreadThick,$fn=8);
              for (a=[0:90:360])
                  rotate(a)
                    translate([Fin.x/2 + ThreadWidth/2,0,(ScrewRecess - ThreadThick)/2])
                      cube(Fin,center=true);
            }

    Which changed the NaN time estimates into actual numbers.

    One key difference may be the small hole in the middle. The four ribs (not two!) now overlap by one thread width around the hole, so they’re not quite coincident and Slic3r produces a tidy model:

    Glass Tile Frame - quad support - Slic3r
    Glass Tile Frame – quad support – Slic3r

    The hole eliminates a smear of infill from the center, which may have something to do with the improvement.

    In any event, I have an improved copypasta recipe for the next screw holes in need of support, even if I don’t understand why it’s better.