glance

Fast CSV inspection for the terminal.

Instant schema inference, SQL-like filtering, interactive scrolling, and multiple output formats. Handles multi-million row files in under a second.

Install

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build
sudo cmake --install build --prefix /usr/local

Requires: C++20 compiler, CMake 3.20+, macOS or Linux.

Usage

glance data.csv                          # table with interactive pager
glance data.csv -n 20                    # first 20 rows
glance data.csv -t 10                    # last 10 rows
glance data.csv --schema                 # inferred types as JSON
glance data.csv --count                  # row count

# Filtering
glance data.csv --where "age > 30"
glance data.csv --where "name contains Al" --where "active == true"
glance data.csv --where "dept == Eng" --where "dept == Sales" --logic or
glance data.csv --where "status == active" -i   # case-insensitive

# Sorting
glance data.csv --sort age
glance data.csv --sort-desc salary

# Column selection
glance data.csv --select "name,age,salary"

# Output formats
glance data.csv --format csv
glance data.csv --format tsv
glance data.csv --format json

# Piping
cat data.csv | glance -
cat data.csv | glance --format json > out.json
glance data.csv --where "age > 30" --format csv > filtered.csv

Example

$ glance employees.csv --where "salary > 90000" --sort-desc salary --select "name,salary,department"
┌─────────┬────────────┬─────────────┐
│ name    │ salary     │ department  │
│ text    │ currency   │ text        │
├─────────┼────────────┼─────────────┤
│ Frank   │ $120000.00 │ Management  │
│ Hank    │ $105000.00 │ Sales       │
│ Charlie │ $95000.00  │ Engineering │
│ Eve     │ $91000.75  │ Engineering │
└─────────┴────────────┴─────────────┘
4 rows | 3 cols | 496 B

Interactive Pager

When output exceeds your terminal height, glance launches an interactive pager automatically (like less). Disable with --no-pager.

Key	Action
j / k / arrows	Scroll up/down
Space / b	Page down/up
g / G	Jump to top/bottom
h / l / left/right	Scroll columns
/	Search
n / N	Next/previous match
q	Quit

Works with piped input too: cat huge.csv | glance - opens the pager reading keys from /dev/tty.

Type Inference

Types are inferred from the first 100 rows, ordered by specificity:

Type	Examples
bool	true, false, yes, no, 0, 1
currency	$12.99, $1,200.00
date	2024-01-15, 01/15/2024
int64	42, -7, +100
float64	3.14, -0.5, 1e10
enum	Low cardinality (< 10% unique)
text	Everything else

Numeric types enable proper numeric sorting and comparison in filters.

Delimiter Detection

Auto-detects: comma, tab, pipe, semicolon. Scores each candidate by count consistency across sampled lines (mean / (1 + stddev)). No flag needed.

Performance

Benchmarked on 5M rows (307 MB), Apple Silicon, Release build:

Operation	Time
Display + schema (lazy)	0.09s
Full filter + sort	0.88s

Key techniques:

• mmap for zero-copy file access
• Lazy parsing: only parse rows needed for display, SIMD-count the rest
• Flat storage: single vector<string_view> with stride, no per-row allocations
• SIMD newline counting: ARM NEON on Apple Silicon / Graviton, SSE2 on x86-64
• memchr fast-path for line-end detection

Options

Usage: glance [file.csv | -] [options]

  -n, --head <N>           Show first N rows (default: 50)
  -t, --tail <N>           Show last N rows
  -s, --schema             Output inferred schema as JSON
  -w, --where <expr>       Filter rows (repeatable)
  -i, --ignore-case        Case-insensitive filtering
  --logic <and|or>         Filter logic (default: and)
  --select <col1,col2,...> Show only specified columns
  --sort <col>             Sort by column (ascending)
  --sort-desc <col>        Sort by column (descending)
  --count                  Output only the count of matching rows
  --format <fmt>           Output format: table, csv, tsv, json
  --no-pager               Disable interactive pager
  -h, --help               Show this help

Filter operators: ==, !=, >, <, >=, <=, contains, starts_with, ends_with