ggcorrplot

• New argument preset for publication-grade output in one token: ggcorrplot(corr, preset = "publication") sets white cell outlines and the colorblind-safe "RdBu" palette. It only fills arguments you did not supply, so anything you pass explicitly (e.g. outline.color, colors, palette) overrides the preset. Defaults to NULL (no preset), leaving existing calls unchanged.
• New argument palette to pick a built-in colorblind-safe diverging palette for the fill gradient: ggcorrplot(corr, palette = "RdBu") (or "PuOr"). Each is an 11-stop ramp with white at zero and the usual cool = negative / warm = positive polarity, so it is a one-word alternative to spelling out colors. Defaults to NULL (use colors), leaving existing calls unchanged.
• New value insig = "stars" to mark the significant cells with significance stars (***/**/* for p <= 0.001/0.01/0.05 -- fixed thresholds, not sig.level) instead of crossing out the insignificant ones. Works without lab, so ggcorrplot(corr, p.mat = p.mat, insig = "stars") is a standalone significance map. The default insig = "pch" is unchanged.
• New arguments lower.method and upper.method for a mixed layout: a different glyph per triangle, e.g. the coefficients as numbers in the lower triangle, circles in the upper, and the variable names on the diagonal (ggcorrplot(corr, lower.method = "number", upper.method = "circle")). Each accepts "square", "circle" or "number"; both default to NULL, leaving the single-method output unchanged (#85). Requested by @Breeze-Hu.
• New argument hc.rect to draw rectangles around the clusters of a hierarchically-ordered correlogram: ggcorrplot(corr, hc.order = TRUE, hc.rect = 3) frames the 3 clusters obtained by cutting the tree. Defaults to NULL (no rectangles). The rectangle outline color is set with the companion argument hc.rect.col (default "gray30", e.g. hc.rect.col = "white").

Main changes

• The plot axis is now always drawn in the matrix (row/column) order. This fixes a bug where hc.order = TRUE was silently ignored for correlation matrices with numeric-looking variable names (e.g. "1", "2", ...): reshape2::melt type-converted such names to a continuous axis sorted by value, discarding the clustering order (#37). As a consequence, the as.is argument no longer affects the plot (it is kept only for backward compatibility) -- a call using as.is = TRUE, which previously produced an alphabetically-ordered axis, now follows the matrix order like every other call. Reported by @cabaez (#37).
• show.diag = FALSE on a non-square matrix no longer blanks the wrong cells. The diagonal has no positional meaning for an m x n matrix, yet the removal wiped min(m, n) cells along the leading diagonal; it now removes only the genuine self-pairs (a row and column naming the same variable), leaving a matrix with disjoint row and column variables untouched. Square matrices, and non-square matrices without dimnames, are unaffected. Follows up the non-square support requested by @mt1022 (#5) and @qalid7 (#10).

Minor changes

• Added an introductory vignette, vignette("ggcorrplot"), walking through the package end to end: the inputs, glyphs and layouts, clustering, coefficient labels, significance styles, and theming.
• Added a second vignette, vignette("publication-ready-correlation-plots"), a gallery of finished publication-ready correlogram recipes (clustered, triangle, significance, circle, colorblind palette, edgeless heatmap, rectangular predictor-by-outcome matrix, and full ggplot2 polish).
• Internal refactor: the data-preparation pipeline and the glyph-layer construction are now factored into internal helpers, with no change to the output of any existing call. This is groundwork for forthcoming per-triangle layout options.
• The help page and the README now link to the Datanovia correlation tutorials for worked examples of the plot and of the p-value matrix.