audrey.feldroy.com

from datetime import datetime
from execnb.nbio import read_nb
from fasthtml.common import *
from fastcore.all import *
from functools import cache
from pathlib import Path, PosixPath

List Live Posts

nbs = L(Path('../arg-blog-fasthtml/nbs').glob('*.ipynb'))
nbs

According to this, I have 34 notebooks in arg-blog-fasthtml/nbs, which matches the 34 cards on audrey.feldroy.com.

Pathlib, User Directory, and PosixPath

PosixPath('~/fun/arg-blog-fasthtml/nbs').expanduser()

To specify the path in terms of my home directory ~, I use PosixPath.

nbs = L(sorted(PosixPath('~/fun/arg-blog-fasthtml/nbs').expanduser().glob('*.ipynb'), reverse=True))
nbs

Here I expand ~ into /Users/arg/, list files that end in .ipynb, and convert the generator object into a readable list with a fastcore L list.

Display the Notebooks List Nicely

for nb in nbs: print(nb.name)

If we just print the filenames, we can see my current approach of naming them with the date and TitleCase title.

New Approach

Doing this has given me insight about how to improve my site:

• Keep naming files as before
• Now for the index page, get the titles from the notebooks instead of the filenames

When there were just a few notebooks, these cards were great. Now I want a more information-dense layout with tighter cards, and with titles containing proper punctuation coming from the notebooks themselves.

Revisit Date Parsing

I'm currently getting dates from ISO 8601-prefixed filenames with this not-great code that I hacked together quickly:

def get_date_from_fname(fname):
    try:
        year, month, day = L(regex.findall(r"\d+", fname))[0:3]
    except Exception:
        year, month, day = 0,0,0
    return f"{year}-{month}-{day}"

I talked with Claude 3.5 Sonnet about it. It generated code that looked awesome at first but wasn't my favorite when I experimented with it carefully. But something good resulted: out of that I learned about datetime.fromisoformat and looked it up in the Python datetime docs.

d = datetime.fromisoformat('2025-01-04')
d

It's actually nice to have datetime objects here because I can get the parts like:

d.year, d.month, d.day

And print them with f-strings:

print(f"{d:%a, %b %-d, %Y}")

I like that combination of readability and abbreviations. I'll try it and see if I still like it later.

Iterate on ISO 8601 Date Parsing

My improved function:

@cache
def get_date_from_iso8601_prefix(fname):
    "Gets date from first 10 chars YYYY-MM-DD of `fname`, where `fname` is like `2025-01-12-Get-Date-From-This.whatever"
    try:
        return datetime.fromisoformat(str(fname)[:10])
    except ValueError: return None

Note: on my first pass writing this notebook, I didn't use the @cache decorator. I waited until the end to cache, to make debugging easier for myself. It'll become clear in the next section on notebook titles why caching is good here.

Now let's test it by grabbing a filename and passing it in.

nbs[0]

nbs[0].name

d = get_date_from_iso8601_prefix(nbs[0].name)
d

Iterate on Getting Notebook Titles

Instead of parsing filenames, I'm going to grab the first cell of each notebook and remove the # prefix.

nbc = read_nb(nbs[0].name)
nbc.cells[0]

{ 'cell_type': 'markdown',

  'id': '7a932161',

  'idx_': 0,

  'metadata': {},

  'source': '# NBClassic Keyboard Shortcuts: Command and Dual-Mode'}

nbc.cells[0].source

nbc.cells[0].source.lstrip('# ')

Make That a Function

def get_title(fname):
    "Get title from `fname` notebook's cell 0 source by stripping '# ' prefix"
    nbc = read_nb(fname)
    return nbc.cells[0].source.lstrip('# ')

get_title(nbs[0])

get_title(nbs[1])

Try It on All Titles

print(nbs.map(get_title))

Find the Broken Titles

def has_newline(x): return '\n' in x
broken = nbs.map(get_title).filter(has_newline)
broken

Fix Broken Titles

broken[0]

broken[0].split('\n')

first(broken[0].split('\n'))

@cache
def get_title(fname):
    "Get title from `fname` notebook's cell 0 source by stripping '# ' prefix"
    nbc = read_nb(fname)
    nbc = nbc.cells[0].source.lstrip('# ')
    if '\n' in nbc:
        return first(nbc.split('\n'))
    return nbc

You can imagine how running this on every notebook would be slow! So we add @cache.

nbs[28]

get_title(nbs[28])

print(nbs.map(get_title))

Iterate on Cards

My cards were created with this FastTag:

@cache
def Card(fname):
    date = get_date_from_fname(fname)
    title = get_title_from_fname(fname)
    style = """
        border: 1px solid #e2e8f0;
        padding: 1.25rem;
        border-radius: 0.5rem;
        background: white;
        box-shadow: 0 1px 3px rgba(0,0,0,0.12);
        transition: transform 0.2s ease;
        cursor: pointer;
        text-decoration: none;
        color: inherit;
        display: block;
    """
    header_style = "margin-bottom: 0.5rem; font-weight: 600;"
    date_style = "color: #666; font-size: 0.875rem;"
    
    c = A(
        Header(H2(title, style=header_style)),
        I(date, style=date_style),
        style=style,
        href=f'/nbs/{fname[4:][:-6]}',
        onmouseover="this.style.transform='translateY(-2px)';this.style.boxShadow='0 4px 6px rgba(0,0,0,0.1)'",
        onmouseout="this.style.transform='none';this.style.boxShadow='0 1px 3px rgba(0,0,0,0.12)'"
    )
    return c

Let's rebuild it from scratch with our new functions.

nbs[0]

get_title(nbs[0])

def Card(fname):
    date = get_date_from_iso8601_prefix(fname.name)
    title = get_title(fname)
    return A(
        Header(H2(title, style="margin:0 0 0.5rem 0;font-size:1.25rem;")),
        Div(f"{date:%a, %b %-d, %Y}", style="font-size: 0.875rem;color:#666;"),
        href=f'/nbs/{fname.name[4:][:-6]}',
        onmouseover="this.style.transform='translateY(-2px)';this.style.boxShadow='0 4px 6px rgba(0,0,0,0.1)'",
        onmouseout="this.style.transform='none';this.style.boxShadow='0 1px 3px rgba(0,0,0,0.12)'",
        style="""border:1px solid #e2e8f0;
        padding:1rem;
        border-radius: 0.5rem;
        background: white;
        box-shadow: 0 1px 3px rgba(0,0,0,0.12);
        transition: transform 0.2s ease;
        cursor: pointer;
        text-decoration: none;
        color: inherit;
        display: block;
    """)
show(Div(Style(':root {font-family: system-ui, -apple-system, "Segoe UI", Roboto, "Helvetica Neue", sans-serif; font-size:16px;} p {line-height: 1.5;}'), Card(nbs[0])))

Here I had to set a root element so rem font sizes would show correctly here in nbclassic, where I'm working from. There's more to get it to look right here, but I'm getting a bit tired.

You may be wondering about how there's a lot of CSS getting repeated in each card instance. There aren't that many cards right now, and this is still way smaller than a React/Tailwind app. I have some interesting ideas here that I'll save for another day.

Bringing Changes Over to My Blog App

My blog app arg-blog-fasthtml has a main.py that isn't notebook-generated.

Note: At this point, considering how many functions I rewrote in this notebook, it would be nice to move that main.py to notebooks. I originally had started writing it in notebooks but had moved to the simple main.py to make troubleshooting deployment on a PaaS easier.

For now, I've updated main.py with all of the above manually. I'm like a manual version of nbdev_export and that tells me that I should automate.

Caching

With @cache on all the functions above, the index page is super snappy locally! Yes, that's the unbounded cache, but I have few and small enough things to cache that I won't worry about it for now. I can explore it another time.

Summary

I've made good progress improving my blog's index page!

1. Better date handling using datetime.fromisoformat() instead of regex parsing
2. Getting proper titles from notebook first cells instead of filenames
3. Tighter, more information-dense cards with improved typography
4. Caching with @cache to keep things snappy

It's working well locally, and we'll see what happens when I deploy. I manually updated my blog app's main.py with these changes, though doing this made me realize I should probably move that code into notebooks and use nbdev_export instead of copying by hand.

There's still room for improvement, but I'm happy with my progress!