The viewer

V opens the file in the viewer (Autoview, F7, shows it next to the list). Binary files open in hex. Files over the size limit show only the beginning. Spreadsheets, CSV, DBF, SQLite and Parquet open as a table (see Table mode).

Searching

As in ZTreeWin: F searches from the start, B backwards from the end, / (or S) from the current page and \ backwards from it; Space continues in the same direction, + and - choose the direction. The match is highlighted. At the search prompt, F2 toggles case sensitivity (ignoring case, accents count too: AÇÃO finds ação) and F4 switches between text and hex (byte pairs such as 4d 5a). A * in the middle of the term matches anything on the same line (error*disk).

Syntax colors

Source code and scripts are colored in about 150 languages (Rust, Python, shell, C, JavaScript, SQL, YAML, Markdown…), recognized by extension, by name (Makefile) or by the first line (#!/bin/bash). The default is syntax_highlight (page 2 of the configuration). Colors stay readable on the current theme’s background, and only the first 20,000 lines are colored, and no line longer than 10,000 characters, so huge files never slow the viewer down.

Rendered Markdown

.md and .markdown files open rendered, also in Autoview, inside archives and compressed (x.md.gz), and the frame says RENDERED. T shows the source at the same place, and comes back; H goes through rendered, text and hex.

You see headings, bold, italic, strikethrough, code, code blocks with syntax colors (C turns them off), nested and numbered lists, task lists ([x]), quotes, rules and tables. Nothing is ever fetched, and no link opens anything:

• A link shows its real address beside its text, as in safe <https://example.com>, so the text cannot pass for another address.
• An image is a line of text: [image] caption <photo.png>.
• A Mermaid diagram is a code block labelled mermaid.
• Raw HTML and HTML comments stay visible as text.
• The YAML block at the start of a file is a code block labelled yaml.

The text always fits the window, so W does nothing here; ← → reach the columns of a table too wide for it. A search looks at the text on the screen, addresses included, and finds a phrase that wraps onto the next row. g and # count the lines of the source. Ctrl+B works on the source’s lines, all of them. & goes to text mode and filters there.

A file over 8 MiB opens as its source, with a note. Rendering also stops at the time limit (3 s, 100 ms in Autoview) or at its memory limit, and the header then says (rendered only the start). A table shows at most 256 columns. Control characters show as ·, including those written as entities like .

Markdown itself hides a few things: link definitions that nothing uses, and table cells beyond the columns of the header. To see everything, T shows the source. Italic and strikethrough depend on the terminal and its font, so they also have a color of their own.

Text direction controls

Some invisible characters change the direction of the text and can make code look different from what it really says (Trojan Source, CVE-2021-42574). The viewer never hides them: each one appears as a red ·, and the header says how many there are.

Text encoding

Xylem finds the encoding by itself, in this order: a decompressed file first, then a BOM (UTF-8, UTF-16 LE or BE), then UTF-16 without a BOM, then UTF-8. A file that is none of these is old text: Xylem guesses CP850 (DOS, Western Europe) or Windows-1252 by counting accented letters. Windows .reg exports are UTF-16 and open as plain text.

O shows the file in the next encoding: UTF-8, CP850, Windows-1252, CP437, UTF-16 LE, UTF-16 BE, and back to UTF-8. The footer says which one, and the header shows its name when it is not UTF-8. If you pick UTF-16 on a file that opened in hex, it switches to text. Your choice stays after you edit the file and come back.

To always use one encoding for old text files instead of the guess, set legacy_charset to cp850, cp1252 or cp437 (K on page 2 of the configuration). The default is auto.

Size limit

The viewer reads only the beginning of a big file, and the header says (showing first N MiB). How much it reads is the viewer_max_size option (L on page 2 of the configuration): auto, 16, 64, 256 or 1024 (MiB). With auto, the default, the limit is 1/32 of the machine’s RAM, never less than 16 MiB and never more than 256 MiB (it is 16 MiB when the RAM cannot be read). The configuration screen shows the value in use, for example auto (256 MiB).

Two cases do not follow it. Autoview always reads 1 MiB. A file on a remote machine (SFTP) is read at most up to 16 MiB, because Esc cannot cancel that read.

A bigger limit costs memory and time: with 1024, an ordinary text file that big takes roughly 2.5 GiB of RAM and about 5 s to open. Very short lines cost more than their bytes, because each line has its own bookkeeping, and so do tabs and hidden direction controls: the viewer keeps lines up to about twice the size limit in memory, and past that the header says (showing first N lines). A log of ordinary lines (about 100 bytes) is never cut by this.

Compressed files

Files ending in .gz, .xz, .zst or .bz2 open decompressed, as text. Xylem looks at the first bytes of the file, not at its name, and opens one level only: a .gz inside a .gz stays compressed. The header says gzip → N bytes (the format and the size after decompression).

Decompression has limits. The viewer reads up to the size limit, takes at most 3 s and uses at most 64 MiB of memory for the xz dictionary or the zstd window. In Autoview the limits are 1 MiB, 100 ms and 16 MiB. When the output or the time limit is hit, the header says (showing first N bytes decompressed). A note tells what happened:

• decompression stopped after 3 s (100 ms in Autoview)
• compressed data truncated or corrupt
• needs more than 64 MiB of memory to decompress (16 MiB in Autoview, with (V to view) when V’s 64 MiB may be enough)

The note goes away at the first key, but the header keeps the problem: (truncated or corrupt), (incomplete: …) when the memory limit stopped decompression partway, or (not decompressed: …) when nothing could be decompressed (the raw bytes are then shown in hex).

Autoview shows the note in its title. In hex mode (H) you see the decompressed bytes. Colors follow the name without the compression extension: app.rs.gz is colored as Rust.

For a .tar.gz (or .tar.xz…) the note says TAR archive: Alt+F5 opens it as a folder.

To see the raw compressed bytes, turn off viewer_decompress (J on page 2 of the configuration).

Table mode

Some files open as a grid instead of text. Xylem goes by the extension, and only for uncompressed files on this disk: a file on a remote machine or inside an archive shows as ordinary contents, with the note Tables are shown only for files on this disk.

The CSV separator (comma, semicolon, tab or vertical bar) is detected from the start of the file.

The frame shows TABLE [2/5: name] when there is more than one sheet or table, and the bottom line where the cursor is: Cell B7 (Row 7 of 120). Under a filter, rows keep the number they have in the file, and this line adds which match it is: Cell A13 (Row 13 of 31, match 2 of 8). The last line of the grid has the whole contents of the cell and the filter in use. Searching (F, B, /) does not look inside the grid and finds nothing there; it works in text mode.

Marks stay under a filter, and a Batch writes what is marked and kept by the filter. They belong to the sheet or table open, and are lost when changing it. If the file changes after marking, the Batch refuses, since other records may be at those places.

The structure

i shows, over the grid, the columns in order with what the file says of each one, like DISPLAY STRUCTURE in dBase and Clipper. The arrows, PgUp, PgDn, Home and End scroll it; Esc, Enter or i close it.

Names, types and defaults come from the file and are made safe like the cells: control and text direction characters show as ·. A .dbf whose header contradicts itself shows ? for the decimals.

The SQL filter

\ or & asks for a condition, like the WHERE of SQL (the word WHERE itself is optional), and shows only the rows that meet it. The prompt starts with the current filter; leave it empty to see all the rows again. A note says Filter matched N rows or SQL error: ...; after an error, the previous filter stays.

age >= 30 AND name LIKE 'Ali%'
WHERE status IN ('open', 'late') OR total > 1000.50
NOT (city = 'Rio' OR city IS NULL)
B ILIKE '%silva%'

Keywords can be in any case.

• Columns: the name of the column, in any case and without quotes, or its letter from A to Z. The letter is the way to reach a name with spaces or symbols. A column that does not exist counts as empty, without an error.
• Values: text in single or double quotes (\' for a quote inside it), numbers, TRUE, FALSE and NULL. Anything in quotes is text, never a column.
• Cells: the type comes from what the cell looks like. Empty is NULL, true and false are booleans, numbers are numbers, and the rest is text. So write code = 7, not code = '7': text and a number are never equal, and a cell with 007 is the number 7. Text comparisons distinguish upper and lower case.
• Size: a filter has at most 1,024 bytes and 128 terms.

A whole query in a database

In a SQLite database, the same prompt takes a whole query, starting with SELECT or WITH, with all that SQLite can do: JOIN, GROUP BY, ORDER BY, functions.

SELECT c.name, sum(o.price) FROM customers c JOIN orders o ON o.code = c.code GROUP BY c.name

The frame says QUERY:, and the grid shows what it gave. A condition then filters those rows, and Ctrl+B writes all that the query gives. Leaving the prompt empty clears the filter and then goes back to the table; Tab, [ and ] go to a table too. A query only reads, one at a time: anything else is refused, and a mistake leaves the screen as it was. The limits below apply, so a slow query stops at 3 s and shows the rows it got; a Batch has no such limit. Xylem does not change databases: UPDATE, DELETE and the like get a message saying so. To change one, use another program: O in the file window opens it in the one set in opener, and '*.db' = 'sqlite3 %f' under [editors] makes E open it in sqlite3.

Limits

A table holds at most 50,000 rows and 256 columns, and takes at most what the viewer’s other limits allow: 3 s (100 ms in Autoview), and a spreadsheet that unpacks to more than the size limit (1 MiB in Autoview) does not open as a table. What lies beyond them does not show and the filter does not see it; the header says so: (showing first 50,000 rows) or (stopped by the time limit: showing N rows). While a table is being read the screen stays free and the note says so; Esc stops the reading and shows what came so far, (stopped by Esc: showing N rows), and the next Esc closes the viewer. A cell shows at most 2,048 characters and then …. Control and text direction characters show as ·.

Tables in C libraries

SQLite databases (.db, .sqlite, .sqlite3) and Parquet files are read by libraries written in C, which do not have Rust’s memory safety. So they open as tables only with V. Autoview reads whatever the cursor passes over, and shows these files as ordinary contents, with the note Parquet or SQLite: V opens the table. Open a database from an untrusted source only if you accept this risk. .ods and .xls files also open as tables only with V: their reader builds the whole sheet first, and that part cannot be stopped.

See also: searching the tagged files.